University of Southern California Dissertations and Theses

The case for moral skepticism

(USC Thesis Other)

The case for moral skepticism

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content

The Case for Moral Skepticism
Matt Lutz

Dissertation Committee:
Stephen Finlay, Chair
Mark Schroeder
Ralph Wedgwood
James Van Cleve
Rachel Walker

Submitted to the Graduate School of the University of Southern California
in partial fulfillment of the requirements for the degree of PhD in Philosophy

Degree Conferred: May 2015
1

Thanks to the faculty and graduate students of the USC Philosophy department. This dissertation
is not just the product of three years of research and writing. It’s the culmination of the countless
debates, conversations, and lessons that made up my time here at USC. Everyone I’ve worked
with has added something to this dissertation in one way or another.

Thanks to my committee. To Jim, for always having an interesting citation to share. To Ralph, for
always pushing back. To Mark, for helping me see the big picture. And to Steve, for making me a
better writer and a better philosopher.

Thanks to my family for always supporting me.
2

TABLE OF CONTENTS
Chapter 1: The Moral Closure Argument 3
Chapter 2: Basic Moral Knowledge by Analogy 31
Chapter 3: Moral Non-Naturalism’s Gettier Problem 59
Chapter 4: What Makes Evolution a Defeater? 86
Chapter 5: Inferential Moral Knowledge and the Is/Ought Gap 117
Chapter 6: The Relevance of Moral Skepticism 142
Bibliography 170

3

Chapter 1: The Moral Closure Argument
1. Knowledge and Skepticism
Skeptical challenges have been a constant feature of the philosophical landscape for as
long as there has been a philosophical landscape. For virtually every proposition of common
sense – that an external world exists, that there is a past, that there is a future, that there are other
minds, that there are any minds, and so on seemingly ad infinitum – there has always been some
irritating skeptic to claim that it is impossible to know that these propositions are true. On some
level, we all feel that these skeptical challenges must be mistaken. We do know some things – in
fact, we know a great many things. When the skeptic challenges us to say how we could possibly
know the things that we claim to know, we frequently attempt to give some account. And this
attempt to answer the skeptic by giving an account of how we can know what we surely do know
is what drives a substantial portion of epistemic theory.
This dialectical situation has become so familiar that it is easy to lose sight of the fact that
not all skeptical challenges are created equal. Some beliefs (or entire discourses) are better
founded than others. Discourse about astrology or phlogiston, for instance, is fundamentally
flawed, and not just for the obvious metaphysical reasons (there is no such thing as phlogiston;
there is no causal connection between astronomical configurations and the events of our
everyday lives), but also for epistemological reasons. The metaphysical results about astrology or
phlogiston teach us that, since these theories are false, all evidence we might have received in
favor of these theories is misleading in some way. But epistemologists shouldn't need an assist
from metaphysicians to tell that the evidence in favor of astrological beliefs was never very good
to begin with. When the epistemologist asks the question “how do you know?” to the astrologer,
he's not merely taking on the irritating role of the skeptic – he's taking the first step in an inquiry
4

that will properly lead to the rejection of astrological beliefs. The reasons that the astrologer will
offer up in response to this skeptical challenge ought to be discovered, ultimately, to be no good
at all. And if the epistemologist finds nothing amiss with the epistemic reasons offered up by the
astrologer, our epistemology is in just as bad a shape as it would be if we could give the skeptic
no account of our knowledge of the external world.
These observations, while mundane, cast the skeptical project in something of a different
light. The mere fact that the skeptic is asking skeptical questions does not entitle us to dismiss
the skeptic out of hand – for as we see in the case of the astrologer, this skeptical insistence is
often the first step on the path to wisdom. Very loosely speaking, there are two kinds of skeptical
challenge – skeptical challenges that annoyingly prod at those who actually do hold knowledge,
and skeptical challenges that reveal that a particular area of discourse lacks the proper epistemic
credentials.
What kind of skeptical challenge does the moral skeptic offer?
The orthodox answer, at least among metaethicists working in the analytic tradition, is
that the moral skeptic belongs to the former camp. We have moral knowledge; the moral skeptic
is little more than a confused (and, perhaps, evil) irritant. But whence this orthodoxy? We hold
moral beliefs very strongly, but is this conviction justified? (Consider: the astrologer holds her
beliefs regarding the import of the location of Jupiter with a very high level of conviction as
well.) A mere display of conviction shouldn't be enough to convince us that moral skepticism
belongs in the category occupied by the merely annoying. The anti-skeptic's job is to disprove
the moral skeptic, or (if disproof is too much to ask) to find some firm ground from which the
moral skeptic's challenge can be resisted.
Discovering the epistemic status of our moral beliefs will be the project of this
5

dissertation. I will argue that we have no moral knowledge. – at the end of the day, the skeptic
wins out. Skeptical concerns, as they relate to ethics, are more like skepticism about astrology
than skepticism about the external world.
1.1 The target refined
Before proceeding, it is important to specify the target of my arguments. I have been
talking about moral beliefs to this point, but this leaves much unspecified. There are many
different accounts of the moral, and not all of them raise the same epistemological issues. For
instance, if someone were to adopt Stevenson's tongue-in-cheek account of the good as being
“pink with yellow trimmings,” their account of the good would have serious defects, but it would
have a tractable epistemology – it is fairly straightforward how we can know that something is
pink with yellow trimmings.
My target in this dissertation will be beliefs in propositions that ascribe mind-independent
non-natural normative properties to objects, actions, events, etc. The fact that my target is non-
natural moral properties should be emphasized. While I think that there are substantial epistemic
problems that confront moral naturalists and non-naturalists alike, the nature of these problems
are very different. Moral naturalists hold that moral properties are natural properties, and are
therefore (roughly speaking) the kinds of properties that we have unproblematic epistemic access
to. The challenge for the moral naturalist is the challenge of locating the moral within the realm
of the natural, of figuring out which of the many natural properties out there are properly referred
to as the moral properties. The challenge for the non-naturalist, on the other hand, is to give an
account of how we can have epistemic access to non-natural properties. It is this latter challenge
that I will be exploring in this dissertation.
Introducing this distinction between natural and non-natural properties raises the question
6

of how this distinction is to be drawn. There may be epistemic problems for “non-natural moral
properties” on one understanding of this term, but not on another understanding of the term. As I
will be using the term here, “non-natural moral properties” refers to moral properties that are (1)
not causally efficacious and (2) neither identical to, constituted by, or constitutive of any natural
properties. (2) follows from (1) given the bridge principle that all and only natural properties are
causally efficacious. I find this bridge principle plausible; those that do not can treat (1) and (2)
as separate commitments.
1

While I have been focusing on moral properties so far (and will continue to do so) rather
than the broader class of normative properties, this is only because the moral is a particularly
interesting and relatively well-understood subset of the normative. My conclusions here might
apply just as well to prudential norms, epistemic norms, or any other kind of norms – but only if
those norms are correctly construed as mind-independent and non-natural. I will not defend any
particular view on the metaphysics of non-moral norms, since this line of inquiry will take us far
away from the questions that are the central concern of this dissertation. But I will say that I am
sympathetic to views on which certain norms, including epistemic norms, are naturalizable even
if moral norms are not. So the scope of my critique here need not generalize to all forms of
normativity.
Thus, my target is non-natural, mind-independent normativity, with a particular interest
in moral facts that are both non-natural and mind-independent. It is facts of this kind that are
vulnerable to a strong skeptical challenge. If moral facts are both mind-independent and non-

1
Several prominent “non-naturalists” fail to count as non-naturalists on this way of drawing the line between
natural and non-natural. For instance, Russ Shafer-Landau explicitly rejects (2) (2003, Ch. 3), and Oddie (2005,
Ch. 7) explicitly rejects (1) (and, I think, rejects (2) as well). So Oddie and Shafer-Landau are not my targets in
this dissertation (except insofar as they argue for views in moral epistemology of which proper non-naturalists
might avail themselves). My targets are, instead, individuals like Moore (1903), Pritchard (1912), Fitzpatrick
(2008), Enoch (2011), Parfit (2014), and Scanlon (2014), who accept (1) and (2).
7

natural, then knowledge of them is impossible. Or so I will argue.
So much for introductory remarks. In this introductory chapter, I will present what I take
to be a very strong argument for moral skepticism – the Moral Closure Argument. This argument
will frame the entire dissertation; defending this argument will be my primary task. After
exploring the Moral Closure argument in detail in Section 2, Section 3 will look at the most
prominent objection to the Moral Closure argument, and in Section 4 I will reject this objection
as mistaken. It will emerge that the question of how to support or object to the Moral Closure
Argument involves taking on some basic methodological questions in epistemology. So the
chapter concludes with a section on methodology in moral epistemology, after which I lay out
the plan for the rest of the dissertation.

2. The Natural World and the Closure Argument
The argument for moral skepticism that I will be offering here is a version of a Closure
Argument. Closure Arguments work by making epistemological hay out of a skeptical
hypothesis. A skeptical hypothesis relative to a certain proposition P is a hypothesis that has two
features. First, in the skeptical hypothesis, P is false. Call this the Falsity Condition. But second,
in a skeptical hypothesis, the world is such a way that it will seem to us that P is true nonetheless.
Call this the Appearance Condition. The evil demon hypothesis and the envatted-brain
hypothesis are familiar skeptical hypotheses (relative to our beliefs about the external world)
from general epistemology. According to the evil demon hypothesis, all of our common-sense
beliefs about the external world are false (the Falsity Condition is satisfied). Yet there is an evil
demon who is using his evil magic powers to make it seem to me as though the world actually is
the way I perceive it to be (so the Appearance Condition is satisfied). And according to the
8

envatted-brain hypothesis, all of our common sense beliefs about the external world are false.
But our brains are hooked up to a computer that generates electrical impulses which stimulate
our brain exactly in the way that sensory inputs from an external world would stimulate our
brain, which of course makes it seem as though the world actually is the way we perceive it to be
(again, note the Falsity and Appearance Conditions).
2

Skeptical hypotheses have long been held to be epistemically troublesome. According to
a line of thought going back at least to Descartes, knowledge requires certainty in the sense that
we must be able to rule out every scenario in which we would be in error about our beliefs.
Skeptical hypotheses are specifically crafted to be scenarios in which we would be in error about
our beliefs, and yet we are unable to know that this scenario does not obtain. The conceivability
of these skeptical hypotheses is then supposed to show that knowledge of many common-sense
beliefs is impossible, because the skeptical hypotheses cannot be ruled out. Within a skeptical
scenario, things will appear to be exactly the same as they appear outside of a skeptical scenario.
Thus, it seems as though we lack any basis for coming to know that we are not in a skeptical
scenario.
Skeptical hypotheses can be deployed in the form of a Closure Argument for skepticism.
This argument schema can be completed by any skeptical hypothesis H relative to a common-
sense proposition P, for all subjects S:
Closure Argument Schema:
1. S does not know that H is false.
2. If S knows that P is true, then S knows that H is false.
3. Therefore, S does not know that P is true.

2
One might wonder here why both the Falsity and Appearance Conditions are needed. Isn't the Appearance
Condition enough to make a hypothesis a skeptical hypothesis? Perhaps so – but a Falsity Condition is needed in
order to preserve the validity of a Closure Argument (more on this shortly).
9

The argument is valid by modus tollens. Filling in the envatted-brain hypothesis for H
and any common-sense proposition about the external world for P yields the result that no one
knows anything about the external world. But to generate a skeptical result about some moral
proposition P, we need to fill in this Closure Argument Schema with a different hypothesis – a
moral skeptical hypothesis.
To get a sense of what a moral skeptical hypothesis would look like, imagine the
following kind of world: In this world, there are no non-natural moral facts.
3
Nothing is morally
right or wrong. Yet in this world, human beings believe that there are moral facts (even though
there are not) because a complicated array of natural facts led people in this world to believe that
there are moral facts. In fact, the natural facts that led to the people in this world having the
moral beliefs that they have are the exact same natural facts that led people in our world to have
moral beliefs.
The story runs something like this (the outlines here are taken from Joyce (2006)):
Human beings, as social animals, are often faced with collective action problems that require
social coordination, with these collective action problems frequently giving individuals the
opportunity to “free ride” on the actions of others. Consequently, groups with certain tendencies,
like a tendency toward altruism coupled with a tendency to punish free-riders, fare better in these
collective action problems, and are therefore capable of surviving better. Thus, the human brain
evolved to push deliberation in the direction of pursuing certain socially-beneficial decisions by,
for instance, making certain alternatives too horrible to contemplate. This proto-moral system of
aversions and attractions gained substance with the introduction of language. Concrete concepts

3
Even those who are tempted to object immediately that moral facts are metaphysically necessary must concede
that it is logically possible for any (non-tautologous) moral proposition to be false. The question at issue is how
we can know this logical possibility to be false. Appeals to metaphysical possibility might seem relevant here,
but, in point of fact, won't take us very far at all. This point will be explored extensively in later chapters.
10

could now be matched to the kinds of actions that triggered certain kinds of aversions, and so we
developed words like “horrifying” to describe those actions that we had strong aversions to
contemplating. This language development allowed the kinds of social information contained in
moral concepts to be efficiently transmitted throughout the society, and it also created a set of
social pressures and standards that shaped the vague pro-social urgings of our primitive brains
into more precise set of social mores for our society. Because the vague pro-social urgings of our
primitive brains are not specific in their content,
4
a large number of sets of social mores are
consistent with these vague urgings, and so every society develops a different – perhaps radically
different! – moral code. Differences in individual socialization and temperament then further
refine these moral codes into the set of moral beliefs that a given individual accepts.
Psychologically, these socialized aversions to certain kinds of actions are indistinguishable from
the moral intuitions of agents in our world.
And once again, it should be emphasized that all the processes here are entirely natural.
Non-natural moral properties do not exist anywhere in this world, and so any beliefs about these
non-natural moral properties would be false.
Call the world imagined here “The Natural World.” In the Natural World nothing is
actually right or wrong. However, social and evolutionary/psychological forces have combined
in this world to make it seem as though there are things that really are right or wrong. As we can
see, the Natural World meets the criteria given above for a skeptical hypothesis.
5
The social and
evolutionary/psychological story provides the Appearance Condition – it tells us why certain
moral beliefs seem true. And we have appended a Falsity Condition – despite the appearances

4
See Sripada (2008) for a defense of this “principles and parameters” model of the innateness of morality.
5
One might object that the Natural World hypothesis is a poor skeptical hypothesis because the truth of the
Appearance Condition entails the falsity of the Falsity Condition – that is, the evolutionary facts appealed to here
are sufficient to guarantee that moral facts do obtain in this world. This is an important objection, and will be
dealt with at length in subsequent chapters, particularly Chapters 2, 5, and 6.
11

generated by our psychological makeup, none of these moral beliefs are true.
6
So, because the
Natural World counts as a skeptical hypothesis relative to our beliefs in non-natural moral
properties, we can fill in our Closure Argument Schema with the Natural World hypothesis,
yielding:
Moral Closure Argument:
1. I do not know that I am not in the Natural World.
2. If I know that killing is wrong, then I know that I am not in the Natural World.
3. Therefore, I do not know that killing is wrong.
The argument generalizes for any subject and for any common-sense moral proposition, and so
yields a general skeptical conclusion about morality.
The story I've told about the Natural World should be rather familiar to those who study
moral psychology, but I am not interested in the Natural World hypothesis as a scientific
hypothesis about the reasons why we form moral beliefs. Instead, I am concerned with exploring
the idea of the Natural World as a skeptical hypothesis. This allows us to think of the Natural
World hypothesis in the context of a Closure Argument, which is a helpful way to think of the
Natural World hypothesis.
When we think of the Natural World as a skeptical hypothesis that can be used in a
Closure Argument, the argument takes on a distinguished pedigree. Epistemologists have puzzled
over Closure Arguments for a very long time indeed, and so know how to make Closure
Arguments that are awfully hard to rebut. The Closure Argument Schema presented earlier in this

6
This might seem like a somewhat strange kind of skeptical hypothesis. After all, the Appearance Condition is
startlingly mundane and uncontroversial – we haven't introduced evil demons or envatted brains, but have filled
in the Appearance Condition from current scientific consensus on the origins of our moral capacities. But the
mundane nature of the scenario on offer does not change the fact that the hypothesis has everything it needs to
count as a skeptical hypothesis. And indeed, the fact that part of this skeptical hypothesis is something that
everyone should already believe in is what makes it a particularly good skeptical hypothesis – or so I will argue
later.
12

section was not formulated randomly – that argument schema and the justifications behind it are
explicitly borrowed from epistemologists concerned with making the most compelling skeptical
argument possible (see Feldman 2003). So in pursuing the Moral Closure Argument, we are
working within a well-established and potentially quite cogent argumentative framework.

3. An Objection
Despite what was just said in the previous paragraph, many moral epistemologists see the
fact that the Moral Closure Argument follows the Closure Argument Schema not as a virtue of
the argument, but instead as a very serious flaw in that argument. A very common line of
response to the Moral Closure Argument says that the argument is doomed to failure from the
beginning because the structure of the argument was borrowed (quite explicitly) from general
epistemological concerns that point in the direction of external world skepticism. But external
world skepticism is a bad view; the closure argument for external world skepticism is unsound.
Many believe that this gives us a reason to think that the Moral Closure Argument is flawed in
some fundamental way. Because we know that the version of the Closure Argument that argues
for external world skepticism is unsound, ethicists can just take whatever strategy is used to
debunk the External World Closure Argument, and apply that to the Moral Closure Argument
with suitable modifications. There's nothing special here, nothing uniquely troubling for ethics.
So it simply does not look promising for someone interested in pursuing a strictly moral
skeptical conclusion

by adopting a Closure Argument (as opposed to a more general skeptical
conclusion, from which moral skepticism may follow a fortiori).
The same concern is supposed to hold for any argument for moral skepticism that is
13

structurally identical to a more general skeptical argument. Hence, Judith Thomson (Harman and
Thomson 1996), in considering one kind of moral skeptical argument claims that “the traditional
argument for moral skepticism lends no more weight to Moral Skepticism than do structurally
similar arguments for skepticism about the physical world, the past, the future, other minds, etc.”
(p. 71), and therefore only finds it fruitful to pursue skeptical arguments with “no structural
analogue for any of the other varieties of local skepticism that we have taken note of” (p. 73).
Writing in a similar vein regarding a handful of different arguments for moral skepticism,
Russ Shafer-Landau (2003) says,
“The existence of global skeptical analogues to these arguments also shows
that we should treat moral and non-moral beliefs as equally (in)vulnerable to
such skepticism, at least until we are presented with further arguments
designed specifically to undermine prospects for moral knowledge” (p. 239-
240).
So, too, Michael Huemer (2005) in defending his version of moral intuitionism against
skeptical concerns:
“We should not consider it a fair move... for someone arguing against ethical
intuitionism to deploy general skeptical arguments... Thus, if some particular
argument against intuitionism can be shown to be merely a special case of a
more general argument impugning our knowledge of those sorts of things,
then I may set that argument aside as not relevant to the current discussion”
(p. 12).
And here's David Enoch:
“It is hard to see why worries about evil demons, or brains in vats, or criteria,
14

or regresses, or fallibility, should prove to be especially problematic with
regard to normative knowledge (or justification, or warrant, or whatever)...
And though I think we should take such skeptical worries very seriously, they
do not give the robust realist any reason to worry that is not shared by
everyone else... If it could be shown that some general skeptical argument is
especially worrying when it comes to normative beliefs, or perhaps with
regard to normative beliefs realistically understood, this would of course
be of interest... But I don't know of any attempt – let alone successful attempt
– to show any such thing” (Enoch 2011, 157-158).
7

These four authors are just a particularly prominent sampling that is representative of a
widespread thought in moral epistemology.
8
The common thread here is that all of these authors
take structural similarities to general skeptical arguments to be a good reason to look elsewhere
when it comes to finding a good skeptical argument.
9
According to these authors, we can know
that the Moral Closure Argument is a bad argument because it’s a special case of the general
Closure Argument Schema. Let's call this objection the Special Case Objection. In the next
section, I will argue that the Special Case Objection is mistaken.

4. How to Refute a Closure Argument
According to the Special Case Objection, the fact that the Moral Closure Argument

7
Framing this dissertation project as a response to this one passage from Enoch would make the issues raised
herein seem to have only a narrow appeal. Nonetheless, one of my main goals is to answer Enoch's implicit
challenge by providing a (successful, I hope) attempt to show what is particularly troubling about moral
skepticism.
8
See also, e.g. Star 2008, McCann 2011, Kulp 2011.
9
I want to bracket the question of whether this is the best interpretation of the objection being raised by each of
these authors. I believe context supports my interpretation here, particularly in the cases of Enoch and
Thompson, both of whom introduce a skeptical challenge to moral knowledge only to dismiss it on the grounds
cited here. At any rate, we can say that many authors have offered up something that looks a lot like the Special
Case Objection, which makes an investigation of that objection a useful endeavor.
15

follows the Closure Argument Schema is a good reason to think that the Moral Closure
Argument is no more successful than other instances of the Schema. In order to see how this
Objection fails, we need to look more closely at the Closure Argument Schema. Again:
Closure Argument Schema:
1. S does not know that H is false.
2. If S knows that P is true, then S knows that H is false.
3. Therefore, S does not know that P is true.
Since the argument is valid, there are two ways that we can go about rejecting this
argument: either an attack on the first premise or an attack on the second premise. (Technically,
they are premise-schema, but we’ll call them premises). Let's focus on the second premise first.
The second premise, as it stands, is in need of justification. The second premise is justified by
appeal to a “Closure Principle” (the principle that gives the Closure Argument its name) that says
that knowledge transfers over known entailment. That is,
Closure: If S knows that P, and P entails Q, and S believes that Q on the basis
of S's belief in P, while retaining knowledge of P throughout his reasoning,
then S knows that Q.
10

A denial of the second premise of the Closure Argument Schema is often taken to amount
to a denial of Closure, but it need not be. Closure is distinct from our second premise.
Specifically, the second premise of the argument assumes that Closure is true, but it also assumes
that all of the auxiliary clauses that make up the antecedent of the conditional claim in Closure
are true; to wit, the second premise incorporates the assumptions that P entails Q, that S believes
that Q on the basis of S's belief that P, and that S retains knowledge of P throughout his

10
An exact formulation of Closure is the subject of some controversy, but something like what is offered here is a
standard formulation. See Hawthorne (2004).
16

reasoning. So our argument turns out to be a little more complicated than a simple modus tollens.
Let's expand the Closure Argument Schema to take this into account:
Expanded Closure Argument Schema:
1. S does not know that H is false.
2. Closure
3. P entails that H is false.
11

4. S (dispositionally) believes that H is false, and does so on the basis of his
belief in P, and S's epistemic situation regarding H or P does not change
during S's reasoning in coming to believe that H is false.
5. Therefore, S does not know that P.
The argument is still valid. Given Closure, if S did know that P, then, given the conditions
specified in 3 and 4, S would know that H is false. But S does not know that H is false – that's
premise 1 of the argument – and so S does not know that P. So if we are to resist the skeptical
conclusion, we need to reject one of 1-4.
It is important to look at the situation in this way because traditional responses to the
Closure Argument – contend that we do know that the skeptical hypothesis is false (reject 1) or
deny Closure (reject 2) – overlook the possibility of rejecting either 3 or 4 in the Expanded
Closure Argument Schema. This is usually seen as an innocent shortcut, since it doesn't look
particularly promising to reject either 3 or 4. But can we make any progress from rejecting 3 or
4?
Let's look at 4 first. 4 is a list of contingent features of a particular agent's doxastic state.
So attacking 4 is a way of attacking the generality of the Closure Argument. There might be

11
The need for this premise shows why the Falsity Condition is a necessary aspect of the formulation of any
skeptical hypothesis.
17

some agents for whom the factors mentioned in 4 don't hold. So we might think that these agents
can avoid the skeptical conclusion. However, this ultimately isn't a very promising way of going
about avoiding the skeptical conclusion. Two observations are salient here. First, a quite large
number of agents are in the doxastic state specified by 4. So while the argument may not be fully
general, it still holds in enough generality to be troubling. Second, the conditions specified in 4
involve only an agent realizing the consequences of his beliefs. Failing to meet these conditions
is thus a kind of epistemic imperfection, and epistemic imperfection is the enemy of knowledge.
An agent who fails to satisfy the conditions in 4 is not in a better position to know a common-
sense proposition than one who does satisfy those conditions. As Gettier reminded us, knowledge
cannot be founded in ignorance.
There is, of course, one condition specified in 4 that is not a kind of epistemic perfection,
i.e. the condition that the agent believes that the skeptical hypothesis is false. But this is
something that all non-skeptics believe, at least dispositionally. So this condition will hold true of
all and only non-skeptics. So this argument won't apply to skeptics – but then again, skeptics
already accept the conclusion of the argument.
Attacking 4 doesn't look promising. So let's look at 3, the premise which says that our
common sense beliefs are incompatible with the skeptical hypothesis. This might seem
unassailable. After all, a skeptical hypothesis, as defined, is one according to which our common
sense beliefs are false. Yet, in practice, 3 might fail to be true if the hypothesis that is being put
forward as a skeptical hypothesis isn't actually a skeptical hypothesis. The philosopher advancing
the closure argument in question might just have poorly chosen her skeptical hypothesis, and
inadvertently come up with a hypothesis that is fully compatible with the truth of our common
sense beliefs. Given the fallibility of philosophers, this is not an idle possibility! In fact, one
18

might view Putnam's (1981) response to the skeptical argument as a way of rejecting 3. Putnam
famously argued that, if the envatted-brain hypothesis were true, the proposition picked out by
our common-sense belief “I have hands” would be one that is fully compatible with the envatted-
brain hypothesis. So we should view challenges to 3 as a live way of attacking the Closure
Argument.
So we can see that a Closure Argument can be attacked in three ways – by rejecting one
of the first, second, or third premises. If the anti-skeptic is to argue that the Closure Argument is
unsound, it will be by rejecting one of these three premises. And, according to the Special Case
Objection, the reasons that we have to reject one of these three premises in the External World
Closure Argument will also give us reason to reject the analogous premise in the Moral Closure
Argument. But this is not the case – the reasons that we might have to reject each of those three
premises are essentially tied up with the subject matter of the External World Closure Argument.
Therefore, the Special Case Objection fails. We can show this by considering each premise in
turn.
First, an anti-skeptic could deny the first premise. That is, we could say that a typical
subject does in fact know that the skeptical hypothesis is false, despite the fact that the
appearances are invariant between the skeptical hypothesis and the common sense hypothesis.
But notice: the first premise is a premise that is essentially about a particular hypothesis and our
epistemic position vis-a-vis that hypothesis. If we want to refute the External World Skeptical
Closure Argument by denying the first premise, we must say what it is about the external world
skeptical hypothesis in question (e.g. the hypothesis that I am a brain in a vat) such that I can
know that this hypothesis is false. Pursuing the same strategy as it relates to the Moral Closure
Argument would be to say what it is about the Natural World Hypothesis such that I can know
19

that this hypothesis is false. But these are two very different tasks, because the skeptical
hypotheses in question are very different. One denies the existence of an external world and
maintains common appearances by reference to a computer simulation. The other denies the
existence of moral facts and maintains common appearances by reference to evolution and social
forces. The contents of these two hypotheses have nothing in common. The two skeptical
hypotheses are similar only in that they are skeptical hypotheses – that is, they both provide
Falsity and Appearance Conditions relative to some set of common sense beliefs. But because
the sets of common sense beliefs in question are so different, their suitability for being positioned
within a Closure Argument is all that these two hypotheses share. So the fact that we can know
that we are not brains in vats gives us no reason whatsoever to believe that we can know that we
are not in the Natural World, and that does not change just because the envatted brain hypothesis
and the Natural World hypothesis can be placed in similar roles in two different arguments.
The second way of refuting the Closure Argument is to deny Closure. This might seem
like a much more promising way of making good on the Special Case Objection. After all, one
might think, if Closure is false, it's false in full generality, and so the Closure Argument is
structurally flawed. There are two problems with pursuing this line too enthusiastically. First,
anti-skeptical positions that proceed by denying Closure are by far the minority view within
epistemology more generally. Closure is highly intuitive, and it is widely held that a denial of
Closure is a cost for any view that does this (see, e.g., Hawthorne (2004)) – there is even a
cottage industry based around taking views that have traditionally denied Closure and modifying
them so that they do not do so, while maintaining their anti-skeptical force (see, e.g., Stine
(1976) or Sosa (1999)). And second, even those who deny Closure do not do so in full generality.
No one (to my knowledge) holds that no version of Closure is true. Among those who deny
20

Closure, the common view is that some restricted version of Closure is true. Dretske, a pioneer
of the Closure-denying response to skeptical arguments, holds that the “S knows that” operator is
“semi-penetrating,” by which he means that agents can sometimes, but not always, come to
know the logical consequences of other things that they know. While Dretske's position
emphasizes that agents cannot always know the consequences of their knowledge, he does still
hold that we can sometimes know the consequences of our knowledge. Dretske is interested in
arguing that we can know that an animal in a cage is a zebra without being in a position to know
that it is not a cleverly-painted mule (as in his famous example), or in arguing that we can know
that we have hands without knowing that we are not envatted. He does not think that it is
impossible to extend knowledge by competent deductive inference. It is uncontroversial that
Closure is true at some level of generality.
To make the Special Case Objection work, then, we will need some reason to think that
the Closure Principle, when properly restricted (if indeed it must be restricted), will fail to
license inferences from “I know that killing is wrong” to “I know that I am not in the Natural
World” if it fails to license inferences from “I know that I have hands” to “I know that I am not a
brain in a vat.” Do we have a reason to think this? That is not an easy question – we will examine
it in more detail in Chapter 6. For now, I am content to note that a failure of Closure when
dealing with external world skeptical hypothesis does nothing by itself to show that Closure will
fail when dealing with a moral skeptical hypothesis. As we've already noted, these two
hypotheses have very different content, and are similar only insofar as they can both be counted
as skeptical hypotheses.
A Closure-denying anti-skeptic might try to claim that this is the decisive point: Closure
fails whenever we are using it to argue for a skeptical conclusion by appeal to a skeptical
21

hypothesis. But this cannot be right. This approach (and the Special Case Objection, generally)
assumes that all instances of the Closure Argument Schema are unsound. But there are some
Closure Arguments that are quite sound. Consider the following proposition – that the Sun
revolves around the Earth as the center of the cosmos. This proposition was widely believed by
many cultures throughout history, and, at the time, had the label of common sense. And consider
the corresponding skeptical hypothesis – that in fact that Earth revolves around the Sun and it is
a simple trick of perspective that makes us think otherwise. Call this skeptical hypothesis the
Copernican Skeptical Hypothesis. Then consider the following argument:
Copernican Closure Argument
1. I do not know that the Copernican Skeptical Hypothesis is false.
2. If I know that the Sun revolves around the Earth as the center of the
cosmos, then I know that the Copernican Skeptical Hypothesis is false.
3. So I do not know that the Sun revolves around the Earth.
This skeptical argument is perfectly sound. And what's more, it's not just the fact that we
live in a modern, post-Copernican era that makes this argument a good one.
12
A contemporary of
Copernicus, having first been exposed to the Copernican theory, but without sufficient evidence
to decide between the new Copernican model and the old Ptolemaic model, should take the
Copernican Closure Argument seriously. Indeed, reflecting on the Copernican Closure Argument
is precisely the kind of thing that should make our Copernican contemporary less sure in her

12
An objection: The Ptolemaic theory is false, and the Copernican theory is true. So isn't the soundness of this
argument explained by the factivity of knowledge rather than any particular claims about Closure or epistemic
warrant? My response: Of course the Ptolemaic theory is false... but so are all non-naturalist realist theories of
normativity, so this doesn't mark out an interesting distinction between the Copernican Closure Argument and the
Moral Closure Argument. Of course that claim will be highly contentious, but I raise it here in order to show why
it is important to bracket considerations relating to truth (to the greatest extent possible) when we are doing
epistemology. The Ptolemaic theory is false – but we know it to be false in virtue of the fact that our evidence
does not support the Ptolemaic theory. In particular, our evidence is insufficient to rule out the competing
Copernican theory, and that is an interesting epistemological point independent of the truth or falsity of the
theories in question.
22

belief in the Ptolemaic model.
Thus, the ability to place the moral skeptical hypothesis within the framework of a
Closure Argument gives us no reason to believe that the resulting Moral Closure Argument is
unsound, without a more general account of when Closure fails, if indeed it does. Closure is true
at least sometimes, and can appear in skeptical arguments that are both cogent and sound. More
can be said about Closure, but I will bracket a more thorough investigation into the generality of
Closure until Chapter 6.
The third way of refuting the Closure Argument is to say that the skeptical hypothesis
being put forward is not really a skeptical hypothesis – the truth of the common-sense
proposition under consideration is fully compatible with the truth of what has been called a
skeptical hypothesis. But once again, this third way of refuting the Closure Argument is
essentially about two propositions and the entailment relationships between them. Perhaps the
envatted-brain skeptical hypothesis does not entail the falsity of “I have hands.” Even if this is
so, that fact would do nothing by itself to show that the Natural World hypothesis does not entail
the falsity of “killing is wrong.” Indeed, we have good reason to think that this is not the case.
We defined the Natural World hypothesis in Section 2 by appending a Falsity Condition to an
evolutionary story about the origins of our moral beliefs. The Natural World hypothesis is, ex
hypothesi, logically inconsistent with the truth of any proposition that entails the existence of
non-natural moral properties.
Therefore, we can see that the Special Case Objection to the Moral Closure Argument is
bound to fail. Since the Closure Argument Schema is valid, if an instance of the Schema is
unsound, this will have to be because one of the premises is false. But each premise in the
Closure Argument Schema contains variables that change depending on the nature of the
23

common sense beliefs under attack and the skeptical hypotheses used to attack them. Thus, we
should expect that there are some ways of filling in the variables in the Closure Argument
Schema which yield all true premises (and thus a sound argument), and there are some ways
which yield a false premise (and thus an unsound argument). So the unsoundness of some
Closure Arguments isn't good evidence that other Closure Arguments – in particular, the Moral
Closure Argument – is itself unsound.
We've seen here that some Closure Arguments are bad – the External World Closure
Argument is, by most estimates – but that some Closure Arguments are good – the Copernican
Closure Argument is. The question now becomes whether the Moral Closure Argument is more
like the External World Closure Argument or more like the Copernican Closure Argument. Of
course, for all I've said so far, the Moral Closure Argument might be the bad kind of Closure
Argument, and one (or more) of the three kinds of anti-skeptical strategies outlined in this
section might serve to refute it. But as should be clear by now, establishing that these anti-
skeptical strategies are successful will require much more than simply gesturing at similarities to
strategies for arguing against external world skepticism. The Special Case Objection goes too
quick. The moral anti-skeptic will have to get her hands dirty.

5. Looking Ahead
Let's briefly take stock. So far, we have looked at a particular hypothesis – the Natural
World hypothesis – that I contend functions as a skeptical hypothesis relative to any of our
common-sense moral beliefs. We also saw how skeptical hypotheses can be used, generally, to
form a skeptical Closure Argument against the common-sense beliefs that the skeptical
hypothesis is a skeptical hypothesis relative to. And putting these two strands together, we
24

developed the Moral Closure Argument for moral skepticism. We saw that the fact that one of the
central thoughts underlying the Moral Closure Argument is a very general skeptical argument
scheme has made many moral epistemologists think that the Moral Closure Argument can be
quickly dismissed. But we also saw that fitting the Moral Closure Argument within a general
skeptical argument scheme created only superficial similarities to more general skeptical
challenges. This is because rejecting any particular premise of the Moral Closure Argument will
involve doing very different things than rejecting the analogous premises in other skeptical
arguments. So while the Moral Closure Argument might well still be rejected, it will take
substantive work to achieve this – merely gesturing at structural similarities won't cut it. Yet
reaching this conclusion is not a victory for the moral skeptic, but merely an invitation to do
more work. The Moral Closure Argument is different from other instances of the Closure
Argument, and it is possible for the Moral Closure Argument to be sound even when other
Closure Arguments are not. But that does not necessarily mean that it is a good argument. The
remainder of the dissertation will be concerned with defending the Moral Closure Argument at
length.
We've seen that three premises of the Expanded Argument are, in principle, open to
attack. But in this case the third premise is unassailable, since the Natural World hypothesis was
formulated specifically so as not to be open to attack. This leaves Closure and the first premise –
that we cannot know the Natural World hypothesis to be false – as the only points in the
argument against which one might push. Closure will be examined in Chapter 6, but since so few
views in epistemology reject (every form of) Closure, my primary task will be to argue that we
cannot know that the Natural World hypothesis is false.
5.1 Methodology and Moorean Maneuvers
25

Since most responses to Closure Arguments involve rejecting the first premise of that
argument, I'll conclude this first chapter with a methodological discussion regarding how the first
premise might be rejected. The first premise of the Moral Closure Argument, recall, says that I
do not know that I am not in the Natural World. Rejecting this premise involves saying that I do
know that I am not in the Natural World. Yet how could I know this?
The most common way to argue that the first premise is false is to take the Moral Closure
Argument and flip it on its head. We do know that killing is wrong. And if we didn't know that
we weren't in the Natural World, we couldn't know that killing is wrong. Therefore, we do know
that we're not in the Natural World. Using a flipped Closure Argument to argue that we can know
a skeptical hypothesis is false is sometimes called a Moorean Maneuver.
A Moorean Argument, if sound, would indeed establish that we know that we are not in
the Natural World. But the Moorean Maneuver certainly seems question-begging. After all, the
first premise in the Moorean Argument is that we do know that killing is wrong, which is exactly
the thing that the skeptic is arguing against. The Moorean is not impressed by this charge of
question-begging: we do know, so the argument is sound, and that's all there is to it (Pryor 2000).
We've barely begun, and it looks like we've already hit an impasse. So we make our way
forward carefully. The anti-skeptic believes that she is entitled to claim that she knows what she
claims to know. The skeptic disagrees. In effect, what's at issue here is a methodological question
about how we should proceed when doing epistemology. We need some place to start from when
we begin our epistemological inquiry, and this impasse demonstrates that the skeptic and the
anti-skeptic likely have very different ideas about what the right starting place is. So we need to
set skepticism and anti-skepticism aside for a moment to engage with a question of methodology
in epistemology.
26

The relevant methodological puzzle here is what as known as the Problem of the
Criterion. Here's Chisholm's explanation of the Problem:
We may distinguish two very general questions. These are, “What do we know?” and
“How are we to decide, in any particular case, whether we know?” The first of these may
also be put by asking, “What is the extent of our knowledge?” and the second, by asking,
“What are the criteria of knowing?”

If we know the answer to either one of these questions, then perhaps we may devise a
procedure that will enable us to answer the other. If we can specify the criteria of
knowledge, we may have a way of deciding how far our knowledge extends. Or if we
know how far it does extend, and are able to say what the things are that we know, then
we may be able to formulate criteria enabling us to mark off the things that we do know
from those that we do not.

But if we do not have the answer to the first question, then, it would seem, we have no
way of answering the second. And if we do not have the answer to the second, then, it
would seem, we have no way of answering the first. (Chisholm 1977, p. 120)

If we are not to sit down and give up on epistemology right off the bat, we need to have
something to say in response to the Problem of the Criterion. We might be “methodists,” who
hold that we should begin by making assumptions about the criteria of our knowledge, and use
that to find out the extent of our knowledge. Or we might be “particularists,” who hold that we
should begin by making assumptions about the extent of our knowledge, and use that to find out
the criteria of our knowledge. Or (this third possibility is frequently overlooked), we might try to
find a kind of wide reflective equilibrium between our intuitions about the criteria of our
knowledge and the extent of our knowledge, giving both sets of intuitions roughly equal weight.
Call this third possibility “reflectivism.”
Now when we look at the initial impasse between the skeptic and the anti-skeptic, an
interpretation of that impasse presents itself. The anti-skeptic might be a particularist, holding
that we should take our intuitions about the extent of our knowledge as our starting point when
doing epistemology. This explains why the anti-skeptic feels entitled to begin with the
27

assumption that killing is wrong: if particularism is the right solution to the Problem of the
Criterion, then assumptions of this kind are entirely legitimate.
13

But particularism cannot be the right methodological approach to take in response to the
Problem of the Criterion. The major problem with particularism is that it does not respect the
factivity of knowledge. If we begin from a starting set of purported knowledge claims that is
inconsistent, a particularist strategy would deem any account of knowledge that required revision
of these inconsistent beliefs to be ipso facto defective: Knowledge claims are of primary
importance, and our standards of knowledge must bend to meet them, no matter how strange a
standard might result. But the existence of any number of paradoxes shows that the set of beliefs
that we intuitively think that we know is inconsistent. And since the particularist approach would
hold that we really do know all of these things that we intuitively think we know, the particularist
would hold that we can know an inconsistent set of propositions. So knowledge must not be
factive.
14

A particularist might object here that I am being too uncharitable. Surely, a particularist
might say, we should be revising our beliefs in light of inconsistencies, discarding the beliefs that
are causing these inconsistencies (and doing so in the way that requires a minimum of revisions
to our starting beliefs). But after doing that, we are entitled to stand firm in our starting beliefs.
This is a reasonable thing to say, to be sure, but it is not particularism. Particularism is the view
according to which particular knowledge claims are taken as being absolutely prior to any
general constraints on knowledge – hence, to reject any particular knowledge claims on the basis

13
And, indeed, anti-skeptics do take this line. See Nelson (2003) for an example of a moral anti-skeptic who makes
this very argument. Nelson responds to Sinnott-Armstrong's (1996) defense of moral skepticism by saying that
all of Sinnott-Armstrong's carefully crafted arguments cannot even begin to get off the ground precisely because
the anti-skeptic can just be a particularist and is thus perfectly free to start from the premise that we know many
moral propositions. Sinnott-Armostrong attempts to evade this objection in his book (2005), but it's not clear
whether he is successful. See DePaul (2009) for a critique.
14
Another possibility is that knowledge is factive, but inconsistent sets of propositions can all be true. But that's
not much more appealing.
28

of a consistency constraint on knowledge is to pursue a methodist or reflectivist strategy. To put
the point more bluntly: the factivity of knowledge is a statement of one of the criteria of
knowledge. So either we take the factivity of knowledge seriously as a starting point in our
epistemological inquiry, or we do not. If we do, then we are taking some of the criteria of
knowledge as our starting point, so we are not particularists. If we do not take the factivity of
knowledge seriously, then if there are any inconsistencies at all in our beliefs before we begin to
engage in epistemic reflection, knowledge will not be factive. There are inconsistencies of this
kind, and it is absurd that knowledge is not factive. Therefore, we ought not be particularists.
Of course, arriving at this conclusion does not mean that there is no way for the anti-
skeptic to answer the skeptic. It just means that one particularly flat-footed way of doing this is
not going to work. The anti-skeptic can still reject the skeptical argument on the grounds that she
knows that killing is wrong. But she can't begin and end with the assumption that she knows
killing is wrong. The need to adopt at least a reflectivist response to the Problem of the Criterion
shows that our claims to knowledge must survive confrontation with epistemic theory. Any view
that says that claims to knowledge can never be threatened by epistemic theory will run into the
problem of the previous paragraph.
The need to adopt at least a reflectivist strategy in response to the Problem of the
Criterion sets the terms of the debate between the moral skeptic and the moral anti-skeptic. An
anti-skeptic must do more than just insist on the existence of moral knowledge. Claims to moral
knowledge must ultimately exist in reflective equilibrium with an account of how our moral
beliefs satisfy all necessary conditions on knowledge. Our reflections on general epistemic
theory must not rule out the possibility of moral knowledge. My task in the rest of the
dissertation will be to show that reflecting on general epistemic theory will show that our beliefs
29

in non-natural moral properties do not have what it takes to count as knowledge.
5.2 The Plan Ahead
The anti-skeptic can give this Moorean response to the Moral Closure Argument only if
she can defend an account of knowledge according to which moral knowledge is possible.
Consequently, in order to defend the first premise of the Moral Closure Argument, the skeptic
must at least demonstrate that the anti-skeptic has no workable model of moral knowledge. This
will be my task over the next four chapters.
In Chapters 2 and 3, I will argue that there is no workable model of basic knowledge
available to the non-naturalist. And in Chapter 5, I will argue that the same is true of inferential
knowledge. Since all knowledge is either basic or inferential (more on this distinction in Chapter
2), this means that there is no workable model of moral knowledge available to the moral non-
naturalist. Along the way, I will argue in Chapter 4 that justified moral belief is rare because of
the presence of a certain defeater for our moral beliefs.
The arguments of Chapters 2 through 5 present a version of a regress argument for
skepticism. (Regress arguments present a skeptical challenge by attempting to rule out any
possible grounds for knowledge within a certain domain.) Thus, I will have provided two
arguments in this dissertation against the possibility of moral knowledge. The first is the Closure
Argument I explored in this chapter. The second is the regress argument. And while the regress
argument and the Closure Argument are distinct, they are mutually-supporting. As Sinnott-
Armstrong notes in his discussion of skeptical arguments, “one crucial premise in [the Moral
Closure Argument] claims that nothing can rule out moral nihilism. The common and the best
way to support that premise is to criticize each method for ruling out moral nihilism. That is just
one instance of what the regress argument does more generally” (Sinnott-Armstrong 2006, p.
30

81).
Thus, while I will present a positive case against the existence of moral knowledge over
the last few chapters, this work will also be in service of a defensive project. The first premise of
the Moral Closure Argument can be rejected if the anti-skeptic can make a plausible claim to
having an account of moral knowledge that would allow here to turn the modus ponens of the
Moral Closure Argument into a modus tollens. I, in turn, will argue that the anti-skeptic can make
no such claim. This defense of the first premise of the Moral Closure Argument, together with
the earlier arguments of this chapter, proves that the first premise of the Moral Closure Argument
is true: the non-naturalist cannot, in any way, know that the Natural World hypothesis is false.
For a non-naturalist trying to resist the Moral Closure Argument, then, she must attack
Premise 2. And Premise 2 will be the subject of Chapter 6. In that chapter, I will look at the idea
that one must not rule out every alternative to a hypothesis in order to know that the hypothesis is
true, but must instead rule out only those alternatives that are “relevant.” This “relevant
alternatives” view can be used to respond to skeptical arguments if it can be shown that the
skeptical hypothesis in the argument is not a “relevant” alternative. In response, I develop a
novel account of what it takes for an alternative to be relevant, and argue that the Natural World
hypothesis is, on this account, a relevant alternative.
31
Chapter 2: Basic Moral Knowledge by Analogy

In the last chapter, we saw that the Moral Closure Argument might be answered by
developing a theory of knowledge according to which many first-order moral propositions are
known. This chapter and the next will be concerned with models of basic knowledge. Chapter 2
will look at attempts to provide a model of basic moral knowledge by borrowing one, wholesale,
from another domain where the possibility of knowledge is relatively uncontroversial. The
general strategy followed by all of these views is to argue that moral knowledge is sufficiently
similar to another kind of basic knowledge that, if basic knowledge of this other domain is
possible, then basic moral knowledge is possible. We will call this kind of strategy Moral
Knowledge by Analogy. This strategy stands in contrast with attempts to craft a model of basic
moral knowledge by appealing to epistemic theory and arguing that no general epistemic
principle rules out the possibility of basic moral knowledge. This second strategy, which we will
call Moral Knowledge by Theory, is the subject of the next chapter.
Section 1 will clarify the matter at hand by explaining what basic moral knowledge is and
why it is important. Sections 2 and 3 will present two different models of basic knowledge on
which we might understand basic moral knowledge – perceptual knowledge and mathematical
knowledge, respectively – and argue that neither of these can serve as a model of moral
knowledge. Section 4 will say what lessons can be learned by the failures of the perceptual and
mathematical models.

1. Basic Moral Knowledge
My thesis in this dissertation is that we have no knowledge of non-natural moral facts. I
am not arguing for this conclusion by arguing that no moral propositions are true (although I do
32
think this), nor am I arguing that we never believe any moral propositions (as non-cognitivists
are wont to do). Rather, it is my contention that we lack moral knowledge because we are
missing something else. In contemporary epistemology, the term “warrant” is used as a bit of
technical terminology that picks out that factor – whatever it is – that must be added to true belief
in order to make that true belief into knowledge. I will be arguing that none of our beliefs in
moral propositions are warranted. (Justification, I will be assuming, is a necessary condition on
warrant, but it is not sufficient for warrant. Accordingly, while justification will be relevant, my
focus is neither exclusively or primarily on how our moral beliefs are justified.) Since warrant is
a necessary condition on knowledge, lack of warrant entails lack of knowledge.
My topic in this chapter is basic knowledge. For any item of knowledge, there will
always be an answer to the question “How do you know that?” The answer to this question will
give the grounds for the agent's believing and subsequently knowing the proposition in question.
If the grounds for the belief include other propositions that the agent herself knows, then the
proposition in question is known inferentially. All other known propositions are known basically;
they are basic knowledge. There are some things that can only be known inferentially by finite
beings like us. We can only have knowledge of the future by testimony or by inference from
observed regularities in the past (the agent must herself know that these regularities have
occurred in order to be able to draw the appropriate inference). On the other hand, it is doubtful
whether there are any propositions that can only be known basically, since it seems that any
proposition can come to be known by testimony,
1
and knowledge by testimony is a kind of
inferential knowledge.
To get a better hold on what kinds of knowledge constitute basic knowledge, it will be
helpful to look at some examples. Knowledge of one's own mental states is a likely example of

1
Possible exceptions include de se knowledge and knowledge of one's own mental states.
33
basic knowledge. When I know that I am sad, it is not because I infer this from some other
proposition that I believe. Rather, this knowledge comes by introspection on my mental states.
Conceptual knowledge is another plausible candidate for basic knowledge. When I know that all
bachelors are unmarried, it is not because I infer this from, e.g., a set of premises having to do
with particular bachelors that I know. Rather, my knowing seems to follow from my
understanding of the concepts “bachelor” and “unmarried,” where “understanding” does not pick
out any known propositions, but something like a cluster of dispositions to think about and use
those concepts in certain ways. Perceptual knowledge is also a candidate for basic knowledge,
but this is controversial. According to some philosophers working in the theory of perception –
sometimes called direct realists – perceptual knowledge is basic knowledge. According to others
– indirect realists – all that we have basic knowledge of is our own mental states. So according to
direct realists, I can have basic knowledge that the cat is on the mat, while according to indirect
realists, I can only have basic knowledge that it seems to me that the cat is on the mat. It follows
that indirect realists hold that we have knowledge of the external world inferentially – typically
by inference to the best explanation from our knowledge about the way things seem to us.
2

Most relevantly, many philosophers believe that we can have basic moral knowledge as
well. That there is basic moral knowledge is even more controversial than that there is basic
knowledge of the external world, but this view is still very popular, and it takes many forms.
Some hold that moral propositions can be known on the basis of our perceptual experiences, i.e.
in the exact same way that we have knowledge of the external world.
3
Some hold that moral

2
Indirect realism and direct realism are not the only two possibilities here. For example, idealists hold that
knowledge of how things seem to me is knowledge of the external world. External world skeptics hold that I
have basic knowledge of the way things seem to me, but I have no inferential knowledge of the external world.
3
See Audi (2013) and McGrath (2004).
34
propositions are known on the basis of our emotional states.
4
Still others hold that moral
propositions are known on the basis of what we might call “pure intuitions,” intellectual
seemings accompanied with (or constituted by) a strong feeling that things could not be
otherwise.
5
While there are many different ways that these views can be spelled out, what they
all have in common is the position that moral propositions are known on the basis of something
other than prior knowledge. Instead, moral knowledge is given to us by our perceptual
experiences, emotions, or intellectual seemings. Thus, all of these views hold that there is basic
moral knowledge. Accordingly, all of these views are my targets in this chapter and the next.

2. Perceptual Moral Knowledge
As we saw in Chapter 1, the skeptic can be answered by developing an independently
acceptable model of moral knowledge. One way of doing this is to say that the same model of
knowledge that works to vindicate some other kind of knowledge can simply be applied (with
minimal modifications) to moral knowledge. Accordingly, the job of the anti-skeptic is not to
develop an account of moral knowledge from the ground up, but instead to show how an
independently motivated model of knowledge is applicable in the moral case. This is the strategy
that we are calling Moral Knowledge by Analogy.
In order to evaluate the prospects for the possibility of basic moral knowledge, we'll
begin by looking at two instances of the Moral Knowledge by Analogy strategy. In this section,
we'll examine the question of whether moral knowledge should be understood as being akin to
perceptual knowledge. I will be assuming that perceptual knowledge can be basic, and so we will

4
This has become an increasingly popular view in recent years, driven by much psychological research showing a
link between the formation of moral judgments and activation in the emotional centers of the brain. For some
representative views, see Oddie (2006) and McCann (2011).
5
This is another very popular view. See Audi (2004), Huemer (2005), and Shafer-Landau (2003) for
representative articulations of this view.
35
be looking at theories of moral knowledge that liken moral knowledge to direct realist theories of
external world knowledge. (We will examine the possibility of whether or not moral knowledge
can be understood as being on par with indirect realist theories of external world knowledge in
Chapter 5.)

2.1 Perceptual Moral Knowledge – Literally
How is moral knowledge like perceptual knowledge? The first view worth examining is
that moral knowledge is perceptual knowledge. This is the view, mentioned in the previous
section, that the grounds for our moral beliefs are exhausted by our perceptual experiences.
When evaluating the prospects for a theory of moral knowledge by perception, a distinction from
Robert Audi is useful. Audi distinguishes between properties that are perceptible and those that
are perceptual. Perceptual properties are any properties that we can know about on the basis of
our sensory experiences. Perceptible properties, on the other hand, are those that our sense
organs put us in causal contact with and which result in some distinctive phenomenology. The
property of being red is perceptible since it is detected by our eyes (for those of us who are not
colorblind) and results in the distinctive visual experience as of something's being red. All
perceptible properties are perceptual, since we can come to know through our sensory
experiences that they are instantiated by attending to the attendant phenomenology that our
senses generate through being in contact with them. But not all perceptual properties need be
perceptible. For an example of a property that is perceptual but not perceptible, Audi offers up
the example of (someone else's) being angry. Anger is a mental state, and it is not the kind of
thing that we can experience directly (when it occurs in others). But anger comes with some
36
characteristic signs – a reddening of the face, a slight trembling, a raising of the voice.
6
These
characteristic signs are all perceptible properties, and we can indeed come to know that someone
is angry by observing these signs. So we can know that someone else is angry through our
sensory experiences, even though the property of being angry is not perceptible. This makes
anger perceptual though not perceptible.
The view that we are now considering says that moral properties are perceptual – but are
they also perceptible? If so, this would make the possibility of basic moral knowledge fairly
straightforward (or at least as straightforward as the possibility of basic knowledge of the
external world by perception). Yet it is impossible for non-natural moral properties to be
perceptible.
7
Perceptible properties are, after all, properties that our sense organs respond to. But
for a non-naturalist, moral properties are not causally efficacious. And because non-natural moral
properties cannot causally affect anything, a fortiori they cannot affect our sense organs and so
are not perceptible.
Given this, the best model for perceptual moral knowledge is one according to which
moral properties are perceptual but not perceptible. This is Robert Audi's view. Audi says that,
like anger, there are a number of perceptible indications of something's being (prima facie)
wrong. Being a killing, for instance, is a perceptible property (our sense organs can respond to
many different aspects of a killing; we see the motion of the knife and hear the scream), and
anything that is a killing is wrong. So while the wrongness of the act is not perceptible, the fact
that it is a killing is perceptible, and so the wrongness of the act is, for Audi, perceptual.

6
Behaviorists about mental states hold that these features are constitutive of anger, so, for a behaviorist, the
property of being angry will be perceptible as well as perceptual. For the purposes of this example, we will
assume that behaviorism is false.
7
I do not think anyone holds that non-natural moral properties are perceptible. Audi, at least, explicitly denies this.
McGrath (2004, 2014), frames the manner in a way that suggests that they might be perceptible as well as
perceptual, but she does not mention the perceptible/perceptual distinction, and does not provide many details for
how moral knowledge by perception is supposed to work.
37
While this may provide us with a workable model of moral knowledge, it is not a model
of basic moral knowledge. In order to have perceptual knowledge of a property that is not
perceptible, the agent in question must have some prior knowledge that links the perceptual
property with the perceptible one. In the case of anger, I must know that a red face, raised voice,
and slight trembling mean that the individual is angry. If I am entirely ignorant of this fact, then I
am not in a position to know that a red-faced, trembling yeller is angry. Accordingly, whenever I
do know by perception that someone is angry, the grounds for my knowing this include my prior
knowledge that red-faced, trembling yellers are angry. And because some prior knowledge of
mine is essentially included in my grounds for knowing that the person before me is angry, this
knowledge is inferential. Importantly, these grounds include having substantive knowledge
about what are good grounds for judging that someone is angry. So if we are to have perceptual
moral knowledge, we must first be in a position to know what are good grounds for judging that
something has a moral property. It follows that Audi's model of perceptual moral knowledge can
only apply if we already have substantive moral knowledge.
None of this is intended to contradict Audi; Audi gladly concedes that his account of
perceptual moral knowledge leads to the conclusion that perceptual moral knowledge is
inferential.
8
He just thinks that he has an account of how we can have prior, basic moral
knowledge. We'll look at this account in Section 3.
To sum up: Because moral properties are not perceptible, and knowledge (by sensory
experience) of properties that are merely perceptual will always end up being a kind of
inferential knowledge, it follows that we have no basic moral knowledge on the basis of our
sensory experiences.

8
The only dispute between Audi and I is terminological. Audi explicitly says that perceptual moral knowledge is
“non-inferential,” but he means something different than I do by “non-inferential.” For Audi, knowledge is non-
inferential whenever it is not arrived at through an explicit process of inferential reasoning.
38

2.2 A More General Analogy
While we do not have basic moral knowledge through our sensory experiences, this does
not entail that basic moral knowledge is not structurally analogous to perceptual knowledge.
Basic knowledge of the external world comes to us through our sensory experiences, so it is
accurate to say that we have knowledge of the external world by our experiences. And these
experiences are experiences as of the properties that I come to know about. For instance, when I
come to know that the apple is red on the basis of my visual perception, it is because I have an
experience as of something being an apple, and an experience as of that apple's being red. On
this basis, I know that the apple is red. The lesson to draw here might be that I can know that
something is the case on the basis of having an experience as of its being the case.
And while I might not be able to have moral knowledge on the basis of a perceptual
experience as of something's being wrong, perhaps I can have moral knowledge on the basis of
some other kind of experience as of something being wrong. For example, Stampe (1987), Oddie
(2005), and McCann (2011) all argue that we have moral knowledge through our desires in
exactly the same way that we have knowledge of the external world through our perceptions.
Oddie uses the slogan that “desires are experiences of value,” saying that desires “stand to values
as ordinary perceptual experiences stand to the objects of perceptual experience...” (Oddie 2005,
p. 40), i.e. objects in the external world. “If there is such a state as the experience of the goodness
of P, then, by analogy with the perceptual case, it would give me a reason to believe that P is
good” (ibid.). For the rest of this section, we will examine the question of whether we can have
moral knowledge on the basis of an “experience of value” (whether or not that “experience” is a
desire).
39
The view that if it seems to S that P, S is prima facie justified
9
in believing P is known as
Phenomenal Conservativism (PC). PC is advanced as part of the dogmatist project in
epistemology. It certainly seems to us as though our basic perceptual beliefs about the external
world are true, and those beliefs are (on the whole) undefeated; the skeptic offers only a “what
if” in the form of a skeptical scenario, and “what if”s aren’t defeaters. Thus can the skeptic be
answered.
10
While PC is a theory of justification, we may be able to extend it enough that it can
serve as a theory of knowledge.
The thought is that while justified true belief falls short of knowledge, it gets us close
enough to knowledge that establishing the justification of certain of our moral beliefs is
tantamount to showing that those beliefs amount to knowledge. After all, the kinds of cases that
Gettier wrote about seem to be unusual. This makes it reasonable to suspect that the additional
factor that must be added to justified true belief in order to obtain knowledge (whatever that
factor might be) is present in the majority of circumstances; hence, showing that a certain class
of beliefs is systematically justified can go a long way toward showing that those beliefs are
typically instances of knowledge. And since PC is an account according to which our moral
beliefs are systematically justified, PC might just have what it takes to show that our moral
beliefs are frequently known.
The basic problem with this approach is that seemings as such are not necessarily reliably
correlated with the truth of the propositions that they represent as true; therefore, believing a
proposition on the basis of its seeming true does not mean that this belief will be reliably formed.

9
Where prima facie justified means justified in the absence of defeaters.
10
Jim Pryor (2000) and Michael Humer (2001) both endorse the view that we can be prima facie justified in our
perceptual experiences, and that this prima facie justification is sufficient to answer the external world skeptic (in
case there are no defeaters), although Huemer is the only one to explicitly defend PC as the source of our prima
facie justification.
40
And since having a reliably formed true belief is a necessary condition on knowledge,
11
the mere
fact that something seems to be true does not ensure that we can have knowledge on the basis of
this seeming. While some particular kinds of seemings, such as perceptual seemings, are reliably
correlated with the truth, it is not perceptual seemings' status as seemings that ensures that they
are so correlated. Consider the following case:
Seeming Pill: A mad neurosurgeon creates a pill, on a whim, that creates the
following disposition in anyone who ingests it: it will seem to this person that
the number of stars in the sky is even whenever he smells freshly baked bread.
The mad neurosurgeon then slips this pill to Harold, who later that day walks
past a bakery. Harold smells the bread, and suddenly it seems to him that the
number of stars in the sky is even. On the basis of this seeming, he forms the
belief that the number of stars in the sky is even.
Suppose the number of stars in the sky really is even. In this case, Harold's belief turns
out to be true, but only by accident. So even if PC is correct and Harold's belief in this case
counts as justified, Harold does not know that the number of stars in the sky is even. So this case
shows that something's merely seeming true isn't the right kind of thing to guarantee that the
proposition that seems true can be known, and this is so even if both the subject possesses no
evidence against the proposition in question and the proposition is true. We might have
knowledge on the basis of something's seeming to be true, but if we do, the causal mechanisms
that generate that seeming are going to matter. Sensory perceptions have the right kinds of causal
mechanisms behind them to ensure that our beliefs that are generated on the basis of our sensory
experiences will, in normal circumstances, be reliable. That's why they can generate knowledge
of the external world. But Harold's seeming in Seeming Pill does not.

11
Although I take it as obvious that reliably formed true belief is not sufficient for knowledge.
41
What Seeming Pill shows is that knowledge requires much more than a proposition's both
seeming to be true and being true. It matters why that proposition seems to be true. Seemings
generated in the wrong kinds of ways may still provide justification, but they cannot be grounds
for knowledge. Accordingly, whether we can have moral knowledge on the basis of something's
seeming true will depend on why moral propositions seem to us to be true, and whether the
explanation of a moral proposition's seeming to be true is the right kind of explanation of the
seeming state. Since our perceptual seemings are the product of sensory perception, while moral
seemings are not, moral seemings cannot be given the same kind of explanation as our perceptual
seemings. So there is at least one important way in which moral knowledge cannot be the same
as perceptual knowledge
To sum up: we attempted to rescue the analogy between moral and perceptual knowledge
by abstracting away from any of the particulars of the mechanisms of perceptual and moral belief
formation, considering the possibility that there might be a fruitful analogy to be found in the
fact that both moral propositions and propositions about the external world will seem true in one
way or another. But as Seeming Pill demonstrates, whether or not a subject has knowledge on the
basis of something's seeming true is going to depend on these very mechanisms that we had
abstracted away from. So while we might be able to understand justification for moral beliefs on
a similar model to perceptual justification, we cannot understand moral knowledge on a similar
model to perceptual knowledge.

3. Mathematical Moral Knowledge
Many will not find the conclusion of the previous section surprising. Knowledge of the
external world is a paradigmatic kind of a posteriori knowledge. But knowledge of basic moral
42
principles is, supposedly, a priori. Hence, if we are trying to pursue the Moral Knowledge by
Analogy strategy, we ought not look to perceptual knowledge to find an easy model for moral
knowledge. Instead, we ought to think of moral facts as being known on an a priori model. So
we turn to math.
The idea that moral knowledge can be understood as being similar to mathematical
knowledge is an old one. It's an idea that can be found in Plato, in Spinoza, in Kant, in W D
Ross, and in many contemporary authors.
12
Some mathematical knowledge is plausibly
construed as basic knowledge (although this isn't uncontroversial – we'll see why shortly), and so
perhaps we can have basic moral knowledge in roughly the same way that we have basic
mathematical knowledge.
However, when it comes to establishing an analogy between mathematical epistemology
and moral epistemology, the question of whether moral knowledge can be understood as being
relevantly similar to mathematical knowledge depends entirely on the nature of mathematical
knowledge. Some accounts of how we have mathematical knowledge might not be the kinds of
things that moral non-naturalists can appeal to. And if this is the case, then the mathematics
analogy will fall apart.
In this section, I will attempt to show that we cannot know moral facts in the same way
that we know mathematical facts. Because there is substantial disagreement on how we do know
mathematical facts, I will proceed piecemeal, by examining the four existing models of
mathematical knowledge and arguing that none of them can also be a model for basic moral
knowledge (setting aside the question of whether or not any of these models is a good model of
mathematical knowledge).

12
See Clark-Doane (forthcoming) for discussion and a helpful list of references.
43
3.1 Reduction and Indispensability
The first two models of mathematical knowledge can be discarded quickly because they
are not models for basic knowledge of non-natural properties. The first strategy attempts to solve
epistemological problems in mathematics by reducing mathematical facts to naturalistically
respectable facts about the world, involving only a nominalistic ontology. This is the nominalist
project in mathematics. On this view, the subject matter of mathematics is not abstract objects,
numbers, but rather the concrete objects in the world. So '2 + 2 = 4' is not a claim that is made
true by the relations between abstract objects, but rather a useful shorthand for describing what
happens when I have a pair of apples and then come into possession of another pair of apples, or
when I have a pair of oranges and then come into possession of another pair of oranges, or any
other situation where groups of the same sizes are being combined.
13
If we can reduce
mathematical facts to facts about apples and the other natural furniture of the world,
14

mathematical epistemology becomes much more tractable. Apples are the kinds of things I can
know about by relatively uncontroversial means.
If this is how we know mathematical properties, then the analogy with mathematics will
not be helpful to the moral non-naturalist. The analogous move in moral epistemology would be
to give a simple account of how we can know about moral properties that can be reduced to
natural properties. But the non-naturalist is committed to there being no properties of this kind.
We will look, then, at views in mathematical epistemology that do not rely on a naturalist
metaphysics.

13
This is a somewhat awkward formulation, but if the nominalist about mathematics is correct, we should be able
to describe these kinds of situations without using the word “two,” thus quantifying over anything in the domain
of numbers. “Pairs” of apples (as I use the word here) are neither numbers nor sets. They are wholly constituted
by an apple and another apple.
14
The actual nature of the reductions that need to be performed is much more complicated than what I've suggested
here, but since none of these details matter for the purposes of our investigation, I have set them aside.
44
The second model of mathematical knowledge is an indispensability model. According to
this view, mathematical facts are indispensable to science. Mathematics is a tool of the scientist,
and scientific theories are typically given in mathematical terms, and confirmed using
mathematical methods. Since science has proved so successful at generating accurate predictions
about the world, this success counts as confirmation of the scientific theories on which these
theories are based. And since mathematics is indispensable for formulating successful scientific
theories, the success of science counts as confirmation of mathematics as well.
This model of mathematical epistemology is available to non-naturalists about
mathematics, so it is the kind of thing that non-naturalists about morality should be able to
appeal to as well. But this is not a model of basic mathematical knowledge. Rather, it is a model
of inferential mathematical knowledge. According to indispensability arguments, mathematical
beliefs are justified along with the scientific theories for which they are indispensable. But
scientific theories are not basic knowledge. They are paradigmatic examples of inferential
knowledge, known by inference to the best explanation from observational data. Similarly, we
should consider indispensability arguments to be arguments for inferential mathematical
knowledge, where mathematical theories are known by inference to the best explanation from
their success in modeling observational data. The manner in which scientific theories are
inferentially known is perhaps slightly different from the manner in which mathematical theories
are inferentially known, but this is of little consequence. Without prior knowledge of
observational data, or knowledge of the success of various scientific theories at explaining this
data, we cannot avail ourselves of indispensability arguments for mathematical knowledge. We
cannot know a priori by reflecting upon the nature of mathematical facts that they are the kinds
of things that would prove indispensable to scientific inquiry.
45
This is not to say that indispensability arguments cannot be used in moral epistemology.
But since this chapter is concerned with the possibility of basic moral knowledge, we will defer a
discussion of whether or not moral facts are indispensable until Chapter 5.

3.2 Conceptual Knowledge
The next model of mathematical knowledge to consider is one on which mathematical
truths are all conceptual truths. This certainly seems to be the case for many of the axioms of
mathematics. For instance, anyone who denies the Successor Axiom (if a is a number, the
successor of a is a number) clearly doesn't understand the concepts of numbers or successors.
More promisingly, Crispin Wright (1983) proved that the axioms of Peano arithmetic can be
derived from the truths of second order-logic, combined with a further principle called Hume's
Principle: The number of Fs = the number of Gs if and only if the Fs and Gs stand in a one-to-
one correspondence. This is the neo-logicist project. While it is doubtful that Hume's Principle
expresses a logical truth, Hume's Principle is plausibly a conceptual truth about the concepts of a
one-to-one correspondence, number, and identity. If Wright is correct, then a substantial portion
of mathematics (Peano arithmetic) can be shown to follow from conceptual truths. This provides
us with a simple route to mathematical knowledge. If mathematical truths are conceptual, we can
know them by reflecting on our mathematical concepts.
If mathematical truths are conceptual truths, and if we are to understand moral knowledge
as being on a par with mathematical knowledge, then it must be the case that moral truths are
also conceptual truths. But the view that moral truths are conceptual truths faces significant
problems stemming from the Open Question Argument.
46
As I understand it, the Open Question Argument runs like this:
15

1. All questions of the form “X is N, but is X M?” are open questions (where N is a natural
predicate and M is a moral predicate).
2. If a question of the form “X is N, but is X M?” is open, then “If X is N, then X is M”
does not express a conceptual truth.
3. Therefore, no claims of the form “If X is N, then X is M” express conceptual truths.
4. If no claims of the form “If X is N, then X is M” express conceptual truths, then there are
no analytic natural-moral property-identity claims.
5. Therefore, there are no analytic natural-moral property-identity claims.
16

Perhaps it is because Moore's conclusion (5) is so well-known and influential that
contemporary advocates of conceptual moral knowledge tend to emphasize that they are not
attempting to provide an analysis of moral concepts. As such, Audi (2008) claims that moral
propositions are self-evident, and an “adequate understanding of a self-evident proposition can
enable us to see its truth by virtue of apprehending conceptual relations” (Audi 2008, p. 486). Yet
he holds that moral propositions are not analytic, saying
“The non-reductivist containment view that I suggest, then, does not
presuppose something that at least Moorean intuitionism would imply is
impossible: a naturalistic analysis of moral (prima facie) reasons... What, then,
can be said of the basis of intuitive knowledge of our apparently self-evident
moral proposition? Does any containment notion apply here? To see how it
might, consider an analogy. Much as we can see that certain items of furniture

15
This is intended as a reconstruction of one popular understanding of the Open Question Argument, not
necessarily as accurate Moore exegesis.
16
Moore thought that this conclusion (5) supported the further conclusion that there are no natural-moral property
identities at all. It is now widely recognized that Moore's argument for this further conclusion fails.
47
are in a room without seeing everything therein, we can see that a certain
concept is contained in another without seeing all that the former contains”
(Audi 2008, p. 479).
Audi's idea here is that it is sufficient for a proposition to pick out a conceptual truth if the right
“concept containment” relation holds between two of the concepts that constitute the proposition,
while an analysis of a concept requires specifying all of what that concept contains. Hence, it
will be possible to have conceptual moral knowledge in virtue of understanding (at least) one
concept-containment relation, even if one is unable to provide a full analysis of moral concepts.
Similarly, Huemer argues that we can have conceptual knowledge, saying that “having a
clear, consistent, and determinate concept is sufficient for one's grasping a universal” and
“adequately grasping a universal cannot cause false intuitions about it.” This explains how we
can have a priori moral knowledge, since “all a priori knowledge is, or derives from, knowledge
of the properties and relations of universals” (Huemer 2005, p. 126, italics in original). Yet he
holds that we do not have analytic moral knowledge, where by “analytic” he means “true by
virtue of the definitions of the relevant concepts,” because “hardly any concepts are definable (in
the philosophers' sense)” (p. 125, my italics). And Cuneo and Shafer-Landau (2014) have
recently defended the view that the “moral fixed points” constitute conceptual truths, where
“moral fixed points” are a set of substantive moral truths which they claim cannot be doubted by
any individual who possesses competence with the relevant concepts. The moral fixed points
include “It is pro tanto wrong to engage in the recreational slaughter of a fellow person” and “it
is pro tanto wrong to break a promise on which another is relying simply for convenience's sake”
(among a host of other claims. See Shafer-Landau and Cuneo, p. 405). As with Huemer and
Audi, Shafer-Landau and Cuneo do not see these moral fixed points as constituting an analysis of
48
moral concepts. They only claim that their truth should be evident to any individual with an
adequate grasp of the concepts involved.
Yet while these views are more modest than views that attempt to provide an analysis of
moral terms, these views still fall victim to the Open Question Argument. For while none of
these views pretends to give an analysis of moral terms, all claim that there are some substantive
moral claims that express conceptual truths, such that any conceptually competent individual will
be able to tell that, say, killing is wrong. Audi believes that Rossian principles that express prima
facie ought claims are conceptual truths, while Cuneo and Shafer-Landau believe that the moral
fixed points are all conceptual truths. Huemer does not build his case around concrete examples,
but does suggest that “enjoyment is good” is a conceptual truth (2005, p. 122). Thus, all of these
views are committed to there being a conceptual truth of the form “If X is N, then X is good” –
e.g. “If X is an instance of recreational slaughter, X is prima facie wrong” or “If X is an instance
of enjoyment, then X is good.” Yet (3) in the Open Question Argument says that there are no
conceptual truths of this kind.
Since (3) follows validly from (1) and (2), a defender of conceptual moral truth must
reject one of these two premises. But before we look at how (1) or (2) might be attacked, we
should note that rejecting the argument to (3) is not costless for the non-naturalist. If (5) is false,
after all, then moral naturalism is true. So the moral non-naturalist needs to be in a position to
support (5), and this has traditionally involved accepting the argument (1)-(5). So moral non-
naturalists who claim that there is conceptual moral knowledge will need another reason to
accept (5). Shafer-Landau and Cuneo attempt to face this problem directly, recommending that
moral non-naturalists pursue a “piecemeal” support for (5), arguing that any particular proposed
analysis of moral terms fails, without committing to the kinds of sweeping generalizations that
49
we see in (1). But since Shafer-Landau and Cuneo are committed to the existence of conceptual
moral truths, it's far from clear what cause they have to be optimistic about the success of such a
piecemeal approach. If there are conceptual moral truths, we need a principled reason for
thinking that none of these truths express identity claims. Without any such reason for optimism,
rejecting (3) cannot be costless for the non-naturalist.
Yet if the non-naturalist is to reject the argument to (3), the most natural premise to reject
here is (1), on the grounds that (1) is far too broad. There are two ways to make this objection to
(1). The first way concedes that every question of the form “X is N, but is X M?” that has ever
been considered has an open feel, but points out that this provides us with, at best, an inductive
argument to the conclusion that all such questions are open. We cannot entirely rule out the
possibility that some future question may be closed. This may be, but defenders of conceptual
moral knowledge will not want to rest their response to Moore on this kind of objection. In
conceding that every question of the form “X is N, but is X M?” that has ever been considered is
open, this objection concedes that we have no actual conceptual moral knowledge (but we could
have conceptual moral knowledge if the right analysis were to be discovered). But defenders of
conceptual moral knowledge want to say that we actually have conceptual moral knowledge, and
so some of the moral claims that we believe express conceptual moral truths. Thus, Shafer-
Landau and Cuneo claim that the “moral fixed points,” like “it is pro tanto wrong to engage in
the recreational slaughter of a fellow human being,” express conceptual truths.
This brings us to the second way to object to (1). We might say that the moral fixed
points or other widely accepted moral propositions do not make for open questions. This
objection holds that “X is an instance of recreational slaughter, but is X wrong?” and other
similar questions will be closed questions.
50
To evaluate this suggestion, we need a clearer idea of what it takes for a question to be
closed. Following Moore (1903), let us say that an open question is one whose answer can
intelligibly be doubted by conceptually competent individuals after reflection on the concepts
involved. It is important that we add the “after reflection” qualifier, since there are some
conceptual truths that are non-obvious. To borrow an example from Robert Audi, “the existence
of a great-grandfather entails the existence of at least four generations” expresses a conceptual
truth, but individuals who are conceptually competent with all of the terms in that sentence need
not agree with it immediately. It's not an obvious truth; seeing that it is true takes a moment of
thinking. But while there are non-obvious conceptual truths, non-obvious conceptual truths have
a way of becoming obvious when conceptually competent individuals reflect on them. Audi's
great-grandfather example certainly has a way of coming to be obvious on reflection.
17

With this in mind, it should be clear that “X is an instance of recreational slaughter, but is
X wrong?” is not a closed question. The claim that all acts of recreational slaughter are wrong
seems like the kind of thing that can be consistently denied by conceptually competent people,
even after reflection. Error theorists, for instance, deny that recreational slaughter is wrong, and I
see no reason to think that error theorists are incompetent with moral concepts. And it is not just
error theorists who deny that recreational slaughter is wrong. There have been a number of
societies who have endorsed the existence of recreational slaughter. Public executions have long
served a role as a grisly form of entertainment, even up through the 19
th
century in England. And
the ancient Romans, with their love of gladiatorial blood sport, clearly did not think that
recreational slaughter was always wrong.
The case of the Ancient Romans introduces a few complications stemming from the fact

17
One might object here that there can be non-obvious conceptual truths that remain non-obvious even after
reflection. But given how I've defined an “open question” here, this would be an objection to the second premise
of the OQA, not the first premise. Accordingly, we will return to this possibility shortly.
51
that the Romans did not speak English. Because Romans spoke Latin rather than English, it's
possible that Roman moral words were governed by concepts distinct from the ones that ground
the use of English moral words; i.e. the English 'good' might not mean the same thing as the
Latin 'bonus.' Hence, it's possible that the Romans could have all failed to be competent with the
moral concept of goodness because they were all competent with some different concept, i.e. the
one picked out by the term 'bonus.' So an advocate of conceptual moral knowledge can avoid the
problems posed by the beliefs of the Romans by holding that Latin words don't have the same
meaning as our English words. But this is a rather remarkable thing to suggest. Linguists, whose
job it is to identify words that have the same meaning, have long translated Latin moral terms
into English. This practice of linguists serves as good evidence that Latin terms pick out the
same concept as English terms. And the fact that linguists have persisted in this practice, even
knowing that the Romans enjoyed recreational slaughter, means that linguists have not taken the
tendency of Romans to view gladiatorial combat as morally permissible to mean that Romans
had different concepts than we do. In general, we will frequently translate the words used by
other cultures to judge actions into the English words 'good' or 'bad,' 'right' or 'wrong' (etc.), and
will do so even if the culture in question has substantially different moral beliefs than our
prevalent in our culture. This is further evidence that it is not a conceptual truth that recreational
slaughter is wrong.
Our standard practice when we talk about Roman citizens and moral philosophers is to
hold that they used genuinely moral terms, and were competent in so doing. So while it is likely
that the Romans – and any other group with a liking for recreational slaughter – are badly
mistaken on the on the substantive question of whether recreational slaughter is wrong, this
distasteful liking does not give us a reason to think that the Romans were similarly mistaken on
52
the conceptual question of what 'wrong' means.
While Cuneo and Shafer-Landau appeal to a set of substantive moral truths to argue that
there are some conceptual moral truths, Audi and Huemer claim that moral truths count as
conceptual truths in virtue of some “conceptual containment” relation. Audi explains conceptual
containment by saying “phenomenologically, one way to explicate the containment idea is to say
that having a thought of a vixen that contains all the elements in the concept—a kind of thought
possible in principle for those who adequately understand the concept—also contains the thought
of femaleness. To have the former thought entails having the latter” (Audi 2008, p. 479). It is
essential to this kind of containment relation that it is necessary conditions that are contained in
the concept that they are necessary for. Hence, being female is contained in the concept of being
a vixen, in Audi's example; to think of something as being a vixen is to think of it as female
(inter alia). Generalizing:
Containment of Necessary Conditions: If it is a conceptual truth that 'all Fs are Gs,'
then to think of something as an F entails thinking of it as a G.
But when we apply Containment of Necessary Conditions to the case of the substantive
moral truths like Audi's mid-level Rossian principles or Cuneo and Shafer-Landau's moral fixed
points, we get the following:
Problematic: If it is a conceptual truth that recreational slaughter is wrong, then to
think of something as an instance of recreational slaughter entails thinking of it as
wrong.
Problematic faces two closely related concerns. First, the central notion in Problematic is
entailment. If one concept contains another, it will be impossible to think of something as falling
under the first notion concept while not thinking of it as falling under the second. As in Audi's
53
example: It's impossible for someone to think of some animal as being a vixen without thinking
of it as being female. Of course, there may be people who will affirm sentences of the form 'x is
a vixen' without affirming sentences of the form 'x is female,' but such an individual must have
some meaning for 'vixen' or 'female' in mind that is different from the typical, conventional
meanings of those words.
Thus, according to Problematic, it will be impossible for anyone to think of something as
being an act of recreational slaughter without thinking of it as being prima facie wrong, so long
as they are using the terms 'recreational slaughter' and 'prima facie wrong' with their
conventional meanings. But this is not impossible. For the non-naturalist, at least, being morally
wrong is something over and above the natural properties; wrongness is something distinct from
any of the natural states of affairs that admit of moral praise or condemnation. Since moral
properties and natural properties have distinct essences, it will be possible to think of a state of
affairs as having one kind of property while lacking the other.
In addition, Problematic is particularly troublesome for the non-naturalist, since if
Problematic is true, then the non-naturalist's main objections to moral naturalism will be
undercut. Problematic says that that thinking of something in descriptive terms entails thinking
of it in normative terms. But the main charges against moral naturalism – Moore's Open
Question Argument, the “normativity objection,”
18
David Enoch's “Just-Too-Different
Intuition”
19
- are all motivated by the same basic thought: to think of something in descriptive
terms entails not thinking of it in normative terms. Once we start describing something in
descriptive terms, the normativity seems to fall out, at the conceptual level. So if Huemer and
Audi are right to say that normative concepts are contained in our descriptive concepts, the Open

18
See, e.g., Dancy (2006, p. 132-142) and Parfit (2011, p. 324-327)
19
Enoch (2011, p. 100-109)
54
Question Argument, “normativity objection,” and “Just-Too-Different Intuition” must be
abandoned. The non-naturalist endorses the existence of conceptual moral truths at the cost of
undercutting the main arguments for non-naturalism.
A defender of conceptual moral truths might also try to attack (2), saying that the open
feel of questions of the form 'X is N, but is X M?' tells us nothing about these questions' status as
conceptual truths, since there are non-obvious conceptual truths. For instance, if mathematical
truths are conceptual truths, then it will be a conceptual truth that the square root of 676 is 26.
But that's not obvious at all. There are two problems with this response. First, as I suggested
previously, non-obvious conceptual truths have a way of becoming obvious after reflection.
Working out a mathematical proof can make the truth of the conclusion of the proof seem
obvious; once the agent working out the proof has seen how the proof works, the conclusion
often gains the phenomenological glow of obviousness. So something's seeming obvious after
reflection is still a hallmark of conceptual truths. Still, there are a number of conceptual truths
that require complex enough proofs that no finite agent could ever hold the whole proof in her
head, and thus the conclusion will never gain the feeling of obviousness. But the defender of
conceptual moral truths won't want to say that moral truths are like this. If it is a conceptual truth
that recreational slaughter is wrong, this is not supposed to be the kind of unobvious conceptual
truth that remains unobvious even after extensive calculation. Quite the opposite! Audi
repeatedly speaks of moral truths as being “self-evident.” Cuneo and Shafer-Landau stress that
they don't think that all conceptual truths are obvious. But while they make a point of saying that
even though the moral fixed points are not “maximally evident,” nonetheless, “nonnaturalists
hold that the degree to which the moral fixed points are evident is quite high” (Cuneo and Shafer-
Landau 2014, p. 16). Indeed, it seems like the best argument to be made in favor of some moral
55
truths being conceptual truths is that certain moral truths seem so obvious.
20
Consequently, the
view that moral truths are non-obvious conceptual truths would undercut the main motivation for
thinking that moral truths are conceptual truths in the first place.
In sum: if we understand the mathematics analogy as arguing that all moral knowledge is
conceptual knowledge, just as in math, the analogy falls apart. People that we have every reason
to think are conceptually competent with moral concepts will deny even paradigmatic moral
truths. We freely translate the words of other languages into words that express moral concepts in
English, even though the other cultures accept very different moral norms. And it's possible to
think of something as having a natural property without thinking of it as having a moral property,
and, indeed, it has seemed clear to many that thinking of something in natural terms involves not
thinking of it in normative terms. For all of these reasons, we should not think that moral truths
are conceptual truths.

3.3 Intuitive Mathematical Knowledge
The last model of mathematical knowledge that we will look at is a pure intuitionist
model. According to this model, mathematical facts are paradigmatic instances of the synthetic a
priori. Mathematical facts are known not by experience or by conceptual analysis; they are just
obvious (in some other way).
If this is what the mathematics analogy amounts to, then the analogy with mathematics
looks like it offers up companions in guilt for morality, not companions in innocence.
Mathematicians shy away from an intuitional mathematical epistemology for many of the same

20
It's remarkable how little argument Audi, Huemer, or Cuneo and Shafer-Landau give for the view that moral
truths are conceptual truths. These authors all state that this is their view, and then show that, if their view is true,
this can solve a number of pressing problems in moral epistemology. But, of course, this is consistent with their
view being false, and problems in moral epistemology being unsolvable. As a friend of mine, Jacob Sparks,
quipped, “Inference to the Most Convenient Explanation” is not a valid argument form.
56
reasons that moral intuitionism is avoided in ethics. The locus classicus for these kinds of
concerns in the philosophy of mathematics is Benacerraf (1973), who argued that we cannot
know about Platonistic mathematical objects on the basis of the fact that there can be no causal
connection between mathematical objects and our beliefs about them. If mathematical properties
are acausal, then how can we have epistemic access to them? How can the truth of our
mathematical beliefs be anything other than an accident? Attempts to solve this problem have
pushed philosophers of mathematics away from a pure intuitionist model of mathematical
knowledge, and towards one of the other models we've looked at in this section. And since non-
natural moral properties are acausal as well, the same concerns arise regarding the possibility of
moral knowledge. Consequently, any attempt to vindicate an intuitional moral epistemology by
analogy with mathematics is not nearly as promising as might have been thought. Intuitionists in
the philosophy of mathematics seem to be vulnerable to the same kinds of objections that
intuitionists in moral philosophy seem to be vulnerable to. Advancing an analogy with
mathematical knowledge is not the end of the problems for intuitionist moral epistemology; it
just makes those same problems arise again in a different domain.

4. Getting Our Hands Dirty
In Section 2, I argued that moral knowledge must be different from perceptual knowledge
because moral beliefs and perceptual beliefs are formed in very different ways. Even if both our
moral belief formation and our perceptual belief formation amount to nothing more than our
endorsing what seems to us to be the case, the fact that our moral seemings and our perceptual
seemings arise through very different processes marks a substantial enough difference between
moral beliefs and perceptual beliefs to call into question the analogy between perceptual
57
knowledge and moral knowledge. In Section 3, I argued that three models of mathematical
knowledge – nominalism, indispensability arguments, and conceptual knowledge – cannot yield
basic knowledge of non-natural normative properties. And a fourth model of mathematical
knowledge – synthetic a priori intuitionism – has problems of its own in the mathematical realm.
This means that there is no model of either mathematical or perceptual knowledge that is both
unproblematic in itself and applicable in the moral domain. Thus, with the two most popular and
promising models of basic knowledge unusable to support the Moral Knowledge by Analogy
strategy, the prospects for this strategy are rather dim.
But the failure of Moral Knowledge by Analogy does not entail the impossibility of basic
moral knowledge. We might think that moral knowledge is unlike other kinds of relatively
unproblematic knowledge, but that it exists nonetheless.
21
Or, we might try to resuscitate the
analogy with mathematics by attempting to solve the Benaceraff problem, and applying this
solution to the moral case. These two strategies ought to be considered together, since they both
involve getting one's hands dirty and diving into epistemic theory. In either case, the anti-skeptic
must argue either that independently plausible epistemic principles prove that we have moral
knowledge, or at least leave open the possibility of moral knowledge. This is the strategy that we
are calling Moral Knowledge by Theory.
In order to make problems for Moral Knowledge by Theory, then, it is incumbent upon
the moral skeptic to make the case that the existence of moral knowledge is impossible. And
since the moral skeptic is not a global skeptic, the impossibility of moral knowledge must derive
somehow from the fact that it is moral knowledge that is impossible. That is, there must be

21
This could either involve showing that moral knowledge is sui generis, or by trying to relate moral knowledge to
another kind of knowledge that is problematic, yet resolving the problems for both the moral knowledge and this
other kind of knowledge. For instance, one could attempt to solve the Benaceraff problem for sythetic a priori
knowledge of Platonic mathematical facts, then apply this solution to vindicate moral knowledge.
58
something about the nature of knowledge, and something about the nature of non-natural moral
properties, that makes knowledge of non-natural moral properties impossible. In the next chapter,
I will argue that this is the case.

59

Chapter 3: Moral Non-Naturalism's Gettier Problem

1. Moral Knowledge by Theory
In Chapter 1, we saw that the moral non-naturalist can resist skeptical concerns about the
possibility of moral knowledge by providing an account of moral knowledge. All knowledge is
either inferential or basic – it is either based on knowledge of other propositions, or it is not. So
if we are to give an account of moral knowledge, we can ask the further question of whether this
account is an account of basic moral knowledge or an account of inferential moral knowledge.
We'll look at models of inferential moral knowledge in Chapter 5. For now, our concern will be
with basic moral knowledge.
In Chapter 2, we looked at attempts to adopt a model of mathematical knowledge or
perceptual knowledge as a model of basic moral knowledge and saw that these attempts were
bound to fail. Moral knowledge is too different from perceptual knowledge for the analogy with
perception to be useful. And while the analogy with mathematics looks more promising, there is
no model of mathematical knowledge which is both available as a model of basic moral
knowledge and which is, in itself, unproblematic. So, simple arguments by analogy to other
varieties of knowledge are not helpful. If we are to provide a model of basic knowledge of non-
natural moral properties, we must do so from scratch. The burden that the anti-skeptic takes on is
the burden of showing that there are no necessary conditions on knowledge that our moral beliefs
cannot satisfy. This is the strategy that we are calling Moral Knowledge by Theory.
In this chapter, I will argue that the Moral Knowledge by Theory strategy is bound to fail:
there is a necessary condition on knowledge that non-natural normative properties cannot satisfy.
In recent work, Mark Nelson (2006) and Walter Sinnott-Armstrong (2006) have both explored
60

the space of possible challenges to moral knowledge, and both have assumed that skepticism
about moral knowledge can take one of three forms: skepticism about moral truth, skepticism
about moral belief, or skepticism about moral justification. But this is not a complete way of
looking at the logical space. Epistemologists have known since Gettier's famous paper that
justified true belief is not sufficient for knowledge. There is at least one other necessary
condition on knowledge: a non-accidentality condition. And this is the condition that basic belief
in non-natural moral properties cannot satisfy. Moral non-naturalism does not have a problem
with justification, truth, or belief; it has a Gettier problem.
My argument for this conclusion will proceed in two stages. First, I will defend a
particular conception of what it takes for a belief to be non-accidentally true. Section 2 will
examine the idea of what a non-accidentality condition is, and argue that such a condition is a
necessary condition on knowledge. Section 3 will then defend a novel account of what is
required for a belief's being non-accidentally true. Section 4 will defend this account against
objections. In the second stage, I will show that this non-accidentality condition cannot be
satisfied by the moral non-naturalist: it is in virtue of being unable to satisfy this previously
unexplored necessary condition on knowledge that basic moral knowledge is impossible. Section
5 will argue that only one possible view can explain how our moral beliefs could be non-
accidentally true, and Section 6 will reject this view on other grounds. The upshot is that basic
knowledge of non-natural moral properties is impossible; if we are to have moral knowledge, it
must be inferential moral knowledge.

61

2. No Accident
One cannot know that P if one's belief that P is merely accidentally true. This thought is
at the center of “No Accident” solutions to the Gettier problem. Thus, according to Unger (1968),
S knows that P if and only if it is not at all an accident that S's belief that P is true. The
sufficiency of a “No Accident” condition on knowledge is controversial, but the necessity of such
a condition seems to be beyond any doubt. If S has a true belief that P, pointing out that S got P
right only by accident counts decisively against claiming that S knows that P.
1
Consequently, if it
is impossible for S to be right about P non-accidentally, it is impossible for S to know that P.

2.1 Non-Accidental Truth, Reliable Methods, and Well-Explained Beliefs
In order to see why it is impossible for moral beliefs to be non-accidentally true, we need
to know more about what it takes for a belief to be true accidentally. We begin by drawing a
distinction between accidentally having a true belief and having a belief that is accidentally true,
since it is only the latter that is destructive to knowledge.
2
I can accidentally have a true belief in
virtue of accidentally finding myself in a position where I am able to use some knowledge-
granting faculty. For instance, I can know (by sight) that my landlord is walking down the street,
even though it is just a matter of luck that my landlord and I happened to be out and about at the
same time and happened to pass each other. Compare this scenario to another: Betsy wants to
decide whether the cat is on the mat, and so she flips a coin, believing that if the coin lands
heads, that will be good evidence that the cat is on the mat. The coin lands heads, and Betsy
forms the belief that the cat is on the mat. In fact, this belief is true. Both Betsy and I accidentally

1
See Yamada (2011).
2
See Setiya (2012, Ch. 3).
62

have true beliefs. But only Betsy does not know the thing that she believes, because only for
Betsy is her belief accidentally true.
Many epistemologists believe that the relevant difference between these two cases is that
it is only in the former case that my belief is reliably formed.
3
I form the belief that that is my
landlord on the basis of a process of visual perception, which, in normal circumstances like mine,
is highly reliable. Betsy is forming her belief on the basis of a coin flip, which is not a reliable
belief-forming process. Cases like this led early reliabilists about knowledge, like David
Armstrong (1973), to conclude that knowledge is reliably-formed true belief (and, thus,
reliability is sufficient for non-accidentality).
However, reliability cannot be sufficient for non-accidentality, as there are cases where an
agent's belief is reliably-formed, yet it is still an accident that his or her belief is true. Consider
the following case from Karl Schafer (2014, p. 12):
The Little Prince: Etienne is an arrogant prince who believes, purely on the
basis of his arrogance, that he is the strongest boy in all of Paris. As it turns
out, Etienne is the strongest boy in Paris, but not because of his exceptional
strength – Etienne is of statistically average strength. Rather, Etienne's father,
the King, who both dotes on his son and is cruel to others, has ordered his
secret police to make sure that there are no boys living in Paris who are
stronger than Etienne – and the secret police have been horrifyingly efficient
in carrying out their orders.

3
Philosophers have analyzed this notion of “reliability” in a number of different ways. For instance, Nozick
(1981) cashed out reliability in terms of sensitivity (If p were not true S would not believe that p) and adherence
(If p were true, S would believe that P), while Sosa (1999) and Williamson (2000) give accounts in terms of
safety (in all nearby worlds in which S believes that p, p is true). I will remain neutral between these accounts
and any number of others here. “Reliable” should be understood as having its intuitive, pre-theoretical meaning.
63

In Etienne's belief is reliably-formed in virtue of the fact that his father sees to it that any
of Etienne's arrogance-based beliefs will come out true. His belief is also Safe – it is true in all
nearby possible worlds where he has that belief.
4
Yet because Etienne is ignorant of the role his
father is playing in seeing to his beliefs' truth, he does not know that he is the strongest boy in
Paris. It is only an accident that Etienne is in circumstances where his arrogance leads him to
have true beliefs.
A focus on the need for reliability is still alive and well in contemporary moral
epistemology. Erik Wielenberg (2010) has recently argued for a “reliabilist moral epistemology”
on the grounds that such a theory can answer the worry that our moral beliefs are only true by
coincidence.
5
Moral propositions are necessarily true, notes Wielenberg, and “where there is no
contingency, there are no coincidences” (p. 461). Wielenberg's motivating idea is that it cannot
be an accident if our moral beliefs are true, since there are no metaphysically possible worlds
where our moral beliefs are false. But it is not at all difficult to construct cases where agents have
beliefs in necessary propositions that are accidentally true. Consider:
Hypnotist: Connie has been hypnotized by a stage magician to form the belief
that 4,294,967,291 is a prime number when she hears a bell ring. (The
hypnotist chose this number at random, just to demonstrate his hypnotic
powers.) Since Connie is strongly susceptible to hypnotism, after she awakes,
she does in fact believe that 4,294,967,291 is prime whenever she hears a bell.

4
Schafer intentionally constructs this case in such a way that Etienne's belief can be made Safe over an
indefinitely large set of possible worlds by indefinitely increasing the efficiency of the secret police. This makes
The Little Prince a counter-example to Safety-based accounts of knowledge.
5
Wielenberg's discussion is couched in terms of “coincidences,” not “accidents,” because the former term is
featured prominently in Sharon Street's “A Darwinian Dilemma for Realist Theories of Value,” to which he is
responding.
64

The hypnotist has given Connie a disposition to form a certain belief in response to a
certain stimulus, and this disposition will reliably result in Connie's forming true beliefs; any
belief that results from this hypnotist-given disposition will be true. This is because
4,294,967,291 is a prime number, so Connie's belief is going to be true in all metaphysically
possible worlds. This makes Connie's reliability different from Etienne's reliability, since the
reliability of Etienne's belief-formation is determined by his contingent circumstances, whereas
Connie's reliability is determined by mathematical necessity. But it is still an accident that
Connie gets things right. In general, then, a disposition to believe a necessarily true proposition
can still result in beliefs that are only accidentally true.
Like The Little Prince, Hypnotist also creates problems for Safety-based accounts of
knowledge. Hypnotist describes a case where an agent uses some method that strongly converges
on a single belief – in all remotely nearby possible worlds, her using this method will result in
the same belief being formed. And also, as a matter of fact, the proposition that she invariably
believes is a necessary proposition. So there is no nearby world where our agent, Connie,
believes something false by use of the method that she is using. But because Connie's method
always delivers a true belief for reasons completely unrelated to the truth of the belief in
question, it is still an accident that her beliefs are true.
Cases like The Little Prince and Hypnotist show that reliability of belief-formation is not
sufficient for a belief's being non-accidentally true. There is a further necessary condition on
non-accidentality that is not met in these two cases. Although Etienne and Connie have formed
true beliefs using reliable methods, they came to use these methods through their use of their
methods is unconnected to those methods' reliability. Some, like David Enoch, take this as an
65

indication that for beliefs to be non-accidentally true, they must be reliably-formed, and there
must be an explanation for this reliability. Here's Enoch:
[V]ery often, when we accept a normative judgment j, it is indeed true that j;
and very often when we do not accept a normative judgments j (or at least
when we reject it), it is indeed false that j. So there is a correlation between
(what the realist takes to be) normative truths and our normative judgments...
And so the realist is committed to an unexplained striking correlation, and this
may just be too much to believe (Enoch 2011, p. 159, my italics).
Enoch calls this need to explain the reliability of our moral beliefs the “strongest version
of the epistemological challenge [to non-naturalist moral realism]” (p. 163, italics in original).
6

While Enoch is correct to say that reliability needs an explanation, not just any
explanation will do. There are, after all, explanations available for the reliability evident in The
Little Prince and Hypnotism. There is an explanation of why Etienne's beliefs regarding his
strength are reliably formed (an explanation in terms of the ruthlessness of the King), and there is
an explanation of why Connie's belief is reliably formed (an explanation that involves the whims
of a hypnotist and the necessity of mathematical facts). Consequently, requiring that our beliefs
be reliable, and that there be some explanation of that reliability, does not make this account of
non-accidentality sufficiently strong to rule out our problem cases.
What we should say is that non-accidentality requires that (i) beliefs are formed by
reliable processes, and (ii) the agent’s adoption of these processes has the right kind of

6
Enoch is explicitly borrowing the idea that some explanation of reliability is needed from Hartry Field (1989).
And Sharon Street (ms) says that moral realists can provide “no reason to think that the causal forces described
by our best scientific explanations shaped our normative judgments in ways that might have led those judgments
to track the truth,” (p. 27, italics in original).
66

explanation: call these Well-Explained Beliefs.
7

2.2 Two Accounts of Well-Explained Belief
What does it take for beliefs to be Well-Explained? The fact that both The Little Prince
and Hypnotism involve the agents finding themselves in fortunate circumstances for reasons
outside of their control looks relevant. The reason why Connie does not know that 4,294,967,291
is a prime number is that someone else is responsible for Connie's adopting the reliable belief-
forming mechanism. And in The Little Prince, the reason that Etienne does not know that he is
the strongest boy in Paris is because someone else is responsible for Etienne's belief being true,
whenever he believes it. This observation motivates Karl Shafer to say that non-accidentally true
belief is just reliably-formed true belief where the agent is responsible for having adopted the
method. Call this the Agential Responsibility view of non-accidentality. Thus, Schafer proposes
the following account of non-accidentally true belief:
1. If a belief is to count as knowledge, it must be reasonable to attribute the
truth of that belief, as well as the positive epistemic features that are relevant
to condition 2 [immediately below], to the believer.
2. If a belief is to count as knowledge, that belief must satisfy some sort of
intuitive “reliability” condition – be that a matter of safety or reliability or
something else less reductive (Schafer 2012, p. 17, italics in original).
Unfortunately for this theory, there are cases where a belief is reliably formed, and this
reliability is attributable to the believer, yet it is still an accident that the agent gets things right.
Consider the following case from Masahiro Yamada (2011):

7
These are intended as necessary, but perhaps not sufficient, conditions on non-accidentality.
67

Two Scales: Luke goes to weigh himself at the gym, and sees two scales
there, a red one and a blue one, and these give conflicting reports of his
weight. Luke flips a coin, and on its basis, decides to trust the blue scale, now
and forever. As it turns out, the blue scale is the perfectly reliable scale, and it
was the red one that is deficient. So Luke always forms correct beliefs about
his weight, because he relies on the blue scale.
In this case, Luke's belief is reliably-formed, and his having this belief is entirely because
of actions that Luke performed. He went to the gym, flipped the coin, picked the blue scale, and
formed the belief.
8
But because the reason that Luke adopted this reliable method is still
essentially arbitrary, it still looks as though Luke's belief is only accidentally true.
9
The coin flip
might have come out differently; Luke might have selected the red scale. So it's possible to have
accidentally true beliefs even if the belief is reliably-formed, and the fact that the belief is
reliably-formed is entirely the responsibility of the believer; invoking notions of agential
involvement won't solve the puzzle.
There is a more natural account of what makes a belief Well-Explained. Think of
Hypnotism. While the fact that the hypnotist is forcing a certain belief-forming mechanism on
Connie looks suspicious, the real problem is not that the hypnotist is responsible for Connie's

8
If we would like to give Luke even more agential responsibility for his belief formation, we can add that Luke
built the scales with his own hands. But if we do this, we should also stipulate that Luke has no idea how to
create a reliable scale. It's a matter of purest luck that he happened to make a scale that is perfectly reliable, and
also a matter of purest luck that it is this reliable scale that he chose to use.
9
One might object that we ought to include Luke's flipping the coin in his method; thus, his method in this case is
unreliable. But as Yamada points out, “the natural way of expressing the situation is that the method Luke is
using to figure out his weight is that of relying on the blue scale. The problem is that he came to rely on the blue
scale through a coin toss. If our account can keep this rather natural distinction between the method one uses and
the way in which one came to use it, that would be preferable” (Yamada 2011, p. 98). The idea here, that I
endorse, is that we should individuate methods in terms of the agent's reasons for belief, not in terms of why the
agent takes those reasons to be good reasons for belief. See Comesaña (2006) for a well-developed account of
method-individuation along these lines.
68

adopting the method that she adopts. The problem is that Connie got her method arbitrarily.
Connie's adoption of this particular belief-forming method has nothing to do with the fact that
this method is reliable. Thus, Kieran Setiya writes:
In the cases we have considered, one's method is reliable but arbitrary. One's
use of this method is random or the product of irrelevant interference. When it
generates true beliefs, they fail to count as knowledge because their truth is
accidental... The problem is clear in the recipe for constructing cases: one's
method is arbitrary in that one's use of it has nothing to do with its being
reliable. In this sense, one's reliability is itself an accident (2012, p. 95-96).
So if an agent is to know a proposition, not only should this agent be using a reliable
belief-forming method, she should be using it because it is reliable. This suggests:
Explanatory Connections View: S's belief that P is Well-Explained if and
only if S's belief that P is the product of a reliable belief-forming method, M,
and there is an explanatory connection between the fact that S is using M and
the fact that M is reliable.
10

The Explanatory Connections View looks like a good account of why Etienne, Connie,
and Luke do not know the true propositions that they believe: all three are using reliable
methods, but none of them are using these methods because their methods are reliable.

3. Explanatory Connections
In this section, we will examine the Explanatory Connections View, and see if a version
of it can be defended. But before doing this, let's briefly take stock of the dialectic thus far. We've

10
This isn't all Setiya has to say on the matter. We'll look at his more detailed proposal shortly.
69

seen that it is a necessary condition on knowledge that our beliefs be non-accidentally true,
which at least involves the belief in the known proposition having been reliably formed. But as
The Little Prince and Hypnotist show us, reliability is not sufficient for non-accidentality. The
belief must be Well-Explained as well – that is, the reliable method, M, must be used for the right
kinds of reasons. Since Well-Explained belief is a necessary condition on non-accidentally true
belief, and non-accidentally true belief is a necessary condition on knowledge, it follows that
Well-Explained belief is a necessary condition on knowledge. As I said at the beginning of this
chapter, basic moral beliefs cannot satisfy the non-accidentality condition on knowledge, which
in turn entails moral skepticism. I can now be more precise about this thesis: basic beliefs in non-
natural moral properties, I will argue, cannot be Well-Explained. And since Well-Explained
belief is a necessary condition on knowledge, if basic moral beliefs cannot be Well-Explained,
basic moral knowledge is impossible. Consequently, we need an account of what it takes for a
belief to be Well-Explained. The Little Prince and Hypnotism suggested two possible accounts of
what it takes for a belief to be Well-Explained: the Agential Responsibility View and the
Explanatory Connections View. Two Scales ruled out the Agential Responsibility View of Well-
Explained belief. So we now turn to the Explanatory Connections view.
3.1 Towards a Workable Version of the Explanatory Connections View
The plausibility of the Explanatory Connections View depends on what we mean by “an
explanatory connection” between S's use of M and the reliability of M. Suppose we mean the
following: an explanatory connection exists between S's use of M and the reliability of M just in
case there is an explanatory chain beginning with S's use of M and ending with the reliability of
M (or vice versa), where the explanations present in this chain can be either complete or partial,
70

so long as the explanans is an ineliminable part of the explanation.
11
Since, on this view, the
explanatory connection between the use of M and the reliability of M must be direct – that is,
there is a single chain that connects use and reliability, not a “common cause” of use and
reliability – we'll call this view 'Non-Accidentality Requires Direct Explanatory Connections'
(NARDEC). So, for instance, Yamada claims that it is a necessary condition on knowledge that
“the truth-conduciveness of M explains why S is using the method M” (Yamada 2011, p. 82).
12

The problem with NARDEC is that it makes trouble for the possibility of inferential
knowledge, particularly knowledge of the future. As Setiya (2012) points out, belief-forming
processes about the future will count as reliable in virtue of how the future facts turn out. But the
future facts cannot explain why I form the beliefs that I form in the present, because there is no
backwards causation from the future to the present.
This puzzle about future knowledge does not mean the end of the Explanatory
Connections view. Perhaps the thing to do is weaken the sense in which the reliability of M and
the use of M must be connected. NARDEC says that reliability must directly explain use, or vice
versa. One weaker view would say that the “explanatory connection” requirement can be
satisfied by an indirect explanatory connection – i.e., two events having a common explanation –
as well as direct explanatory connections. So according to this weaker view, if some common

11
By “ineliminable,” I mean that if we were to modify the explanation by removing or changing the fact in
question, doing so would reduce the explanatory power of the explanans. This “ineliminability” clause is
intended to rule out cases where extraneous facts are conjoined with good explanations, thereby creating a new
explanation of which the extraneous fact is a “part.” The “ineliminable” clause is redundant on some accounts of
explanation – I include it here for the sake of clarity and completeness.
12
Notice that Yamada does not consider explanatory chains going from use to reliability; he only considers the case
where reliability explains use. For the most part, I will follow Yamada in doing this, since cases where a belief is
Well-Explained because the use of a reliable method explains that method's reliability are few and far between. If
there are cases where the explanation does run in the other direction, then, according to NARDEC, the beliefs in
question will be Well-Explained, but the rest of this section makes the simplifying assumption that explanation
always runs from reliability to use. This simplifying assumption will not affect the overall conclusions of this
paper; I argue in Section 5.4 that basic moral beliefs cannot be Well-Explained by use explaining reliability.
71

factor explains both the reliability of M and the use of M, there is an appropriate explanatory
connection between reliability and use. Call this view 'Non-Accidentality Requires Indirect
Explanatory Connections' (NARIEC). NARIEC explains how we can have knowledge of the
future. When it comes to the problem of future knowledge, “the solution is to find a common
cause. Past and future reliability are explained by the same laws of nature” (Setiya 2012, p. 107).
Thus, there is a common cause – the laws of nature – that explains both why we reason
inductively, as well as why our method of inductive reasoning is reliable. Setiya argues for
NARIEC as a better account of Well-Explained belief than NARDEC on these grounds.

And
Elizabeth Tropman seems to endorse a version of NARIEC when she writes “if the wrongness of
cheating does nothing to help explain, even indirectly, my belief that this conduct is wrong,
skeptics may be correct to say that I do not know it to be wrong” (Tropman 2014, p. 137,
emphasis mine).
But NARIEC is also a poor account of Well-Explained belief. For consider the following
case:
Botched Brain Transplant: Last night, while Keenan slept, aliens crept into
his room, cut open his head, removed his brain, and placed it in a vat. When
he awoke the next morning, it seemed as though he was still in his room,
when in actuality he was in an alien spaceship light-years away. But the alien
neurosurgeon was a little clumsy, and damaged Keenan's brain in the
transplant process. This damage had only one effect on Keenan's cognitive
functioning: he is now disposed to disbelieve everything that seems to him to
be true through his senses.
72

Since Keenan is in a vat that is feeding him systematically false seemings, Keenan's
method of disbelieving everything that he sees is going to be highly reliable. And there is an
indirect explanatory connection linking the fact that this method is reliable and the fact that
Keenan is using this method: both facts were caused by the botched brain transplant. But it is just
an accident that Keenan is using a method that is reliable, so his belief is not Well-Explained.
Botched Brain Transplant might seem like a particularly remote case, but counter-
examples to the sufficiency of NARIEC as an account of Well-Explained belief are easy to come
by. For instance, every single physical state of affairs in this universe has a common cause – the
Big Bang. So if all that is required for Well-Explained belief is locating a common explanatory
nexus between formation of beliefs and reliability of methods, all beliefs regarding physical
states of the world will be Well-Explained, since a causal nexus can be located in the Big Bang.
This makes all our previous cases counter-examples to NARIEC as well. The more general point
is that indirect explanatory connections are far too common. Common partial explanations of
otherwise unrelated phenomena are easy to find if one looks back far enough. And trying to place
a limit on how far back explanatory connections can extend is both ad hoc and intrinsically
problematic. We've known since Zeno that the simple act of firing a bow at a target involves an
infinite number of steps that are directly causally connected. NARIEC is just too weak to provide
sufficiency conditions for Well-Explained belief.
3.2 The Crucial Point
To sum up: The Explanatory Connections View is highly intuitive, and denying it will
open up the possibility for accidentally true belief of the kind evident in Two Scales. Refining the
Explanatory Connections view with a commitment to NARDEC makes knowledge of the future
73

impossible. But the weaker NARIEC fails to rule out the possibility of accidentally true belief of
the kind evident in Botched Brain Transplant and a huge variety of other cases. So we're faced
with a puzzle.
While the puzzle we face here may seem as though it concerns a remote side-issue in
epistemic theory (a necessary condition on a necessary condition on knowledge puts us deep in
the morass of the Gettier problem), the larger question of whether or not moral non-naturalism
has a Gettier problem hinges on what is required for non-accidentality. So, for example, Street
(ms) argues that moral realism has an epistemic problem because there is no explanation of why
our moral beliefs could be reliably-formed. But as Enoch (2011) has pointed out in reply, the
phenomenon of moral supervenience shows us that there are explanations of why our moral
beliefs could be reliably-formed: they are just indirect explanations. And Setiya (2013) argues
that a lack of explanatory connections between the moral facts and our beliefs is the fundamental
epistemic problem for moral realism. But Setiya accepts NARIEC – he thinks that indirect
explanatory connections between the reliability of a method and the use of that method are
sufficient for non-accidentality. And Enoch has showed that indirect explanatory connections can
exist between reliability and use. So neither Street nor Setiya has identified a real epistemic
problem for the moral non-naturalist. Yet the non-naturalist is not yet in the clear, since as we've
just seen, NARIEC is too weak to provide sufficient conditions for Well-Explained Belief.
We need an account of Well-Explained Belief that is stronger than NARIEC but weaker
than NARDEC. And whether or not basic moral knowledge exists will turn on what this account
is. So how do we solve the puzzle?
The key to solving this puzzle rests on realizing that indirect connections are only
74

required to secure inferential knowledge, such as knowledge of the future. But if we set
inferential knowledge to the side, then the pressure to accept indirect connections disappears. If
we are looking only at basic knowledge, as we are in this chapter, then a version of NARDEC
looks correct. That is,
Basic Non-Accidentality Requires Direct Explanatory Connections
(BNARDEC): S's basic belief in P is Well-Explained if and only if S's belief
that P is formed through a reliable method M, and there is a direct explanatory
connection between the fact that M is reliable and the fact that S uses M.
Since direct explanatory connections between a method's reliability and its use are
necessary and sufficient for a basic belief being Well-Explained, and are therefore necessary for
the non-accidentality of basic knowledge, BNARDEC entails
Basic Non-Accidentality is Direct (BNAD): If S has basic knowledge that P
through method M, then M is reliable and there is a direct explanatory
connection between the fact that M is reliable and the fact that S uses M.

4. Just Basic Non-Accidentality? Two Objections
We've solved the puzzle of how to account for Well-Explained Belief by restricting our
focus to the non-accidentality of basic beliefs. While this provides a principle that is weaker than
NARDEC but stronger than NARIEC and accounts for all of our data, this restriction might seem
problematic for other reasons. In this section, I'll briefly look at why one might be concerned
about this kind of scope restriction, and argue that there is no great cause for alarm. In fact, we
have very good reason to treat basic knowledge and inferential knowledge differently from the
75

standpoint of non-accidentality.
First, one might worry that giving a necessary condition on non-accidentality that applies
only to basic knowledge will result in an unacceptably disjunctive account of non-accidentality.
If both basic and inferential knowledge involve non-accidentally true belief, it may look like a
problem if basic beliefs are non-accidentally true in a different way than inferential beliefs. But
there is no problem here – in fact, this is what we should expect. The way in which our beliefs
relate to the facts in order for our beliefs to count as knowledge cannot be independent of our
grounds for those beliefs. As Williamson has proved, knowledge is prime: we cannot factor
knowledge into independent internal and external conditions.
13
This means that there must be
some connection between the internal aspects of knowledge (our belief and our grounds for
belief) and the external aspects of knowledge (the truth of that belief and the reliable connection
between the belief and the truth).
14
So we should expect that basic knowledge and inferential
knowledge involve different kinds of connection to the truth, in virtue of the fact that the grounds
for basic knowledge and inferential knowledge are different. It makes sense that inferential

13
Williamson (2000, Ch. 3). For those unfamiliar with Williamson's argument, here it is in brief: Suppose I see
water out of my left eye, while there is a hologram of water positioned in front of my right eye. However, I have
a brain lesion of some sort that prevents me from registering any data from my right eye. Then take another case
where I see water out of my right eye, while there is a hologram of water in front of my left eye; yet in this case,
it is my left eye that is lesion-impaired. In each of these cases, it seems that I know (by sight) that there is water
in front of me. Yet now consider a third case, in which the internal condition from the first case is matched with
the external condition from the second case (or vice versa). In such a case, I have a visual experience as of water,
with this visual experience generated by the input to my left eye (and only my left eye). Yet what is in front of
my left eye is a hologram of water; the actual water is in front of my right eye, which I cannot see out of. In such
a case, I do not know that there is water in front of me.
This case shows that knowledge is not just a conjunction of some purely internal condition and some purely
external condition. Of course, we can think about the purely internal and purely external aspects of a state of
knowledge, so knowledge is in that sense factorable. But any such division will sacrifice something essential to
what makes that state a state of knowledge. Knowledge is more than just an internal factor conjoined with an
external factor. The internal factor and external factor need to line up in the right kind of way. If this is not
required, then any internal state of knowledge should be conjoinable with any external state of knowledge of the
same proposition, and that any such conjunction will also count as knowledge. But Williamson's argument shows
us that this free recombination of internal and external conditions will not always count as knowledge.
14
See Schroeder (forthcoming) for similar thoughts.
76

knowledge will sometimes involve only an indirect connection with the truth, since inferential
knowledge is itself indirect, mediated as it is by prior knowledge of other propositions (e.g., our
knowledge of the future is mediated by our knowledge of the past). On the other hand, basic
knowledge is obtained by standing in some direct relation (e.g. perception) to the fact itself.
Accordingly, we should expect the reliability of the subject's belief-forming method to be
directly explained. So there is a rather natural reason to think that the reliability of inferential
beliefs and basic beliefs will be explained in different ways.
Distinguishing between the non-accidentality conditions for basic and inferential
knowledge also helps with the problem of how to sort the unproblematic indirect explanatory
connections (e.g. the indirect explanatory connections that are essential to typical knowledge of
the future) from the much larger class of problematic indirect explanatory connections (e.g. the
indirect explanatory connections that exist between everything physical, via the Big Bang). It
does this by, in effect, imposing a cognitive access requirement on the indirect explanatory
connections between method and reliability. What is required for knowledge of the future (e.g.)
is that the agent in question has knowledge of the past or the laws of nature in such a way that he
or she might justifiably infer facts about the future from them.
It would also not be accurate to accuse the model suggested here of being disjunctive.
Rather, what I am proposing here is a recursive model of knowledge, where basic knowledge
constitutes the “base case,” and (a version of) Closure gives us the “recursion step,” allowing us
to build inferential knowledge out of our basic knowledge. Thus, BNAD constitutes a necessary
condition on the base case in a recursive account of knowledge. While there is much more that
could be said on the subject of a recursive account of knowledge, developing this model of
77

knowledge will take us too far from issues related to moral skepticism, so I will not pursue it
here. I mention the recursive account only to show that BNAD need not lead us to accept an
ugly, disjunctive account of knowledge.
One might also worry that restricting the account in this way to only talk about basic
knowledge will make inferential knowledge strictly easier to come by than basic knowledge,
since there is a necessary condition on knowledge that applies only to basic knowledge.
15
But
this worry is misplaced. Inferential knowledge is had on the basis of propositions that are
themselves known. Because non-accidentality is a necessary condition on knowledge, the
premises on which the inferential knowledge is grounded will be non-accidentally true. Since
proper use of inference in gaining inferential knowledge is knowledge-transmitting (i.e.
knowledge of the premises is transferred to knowledge of the conclusion), proper use of
inference in gaining inferential knowledge must be non-accidentality-transmitting as well. Of
course, inference is not always knowledge-transmitting, and giving a full account of when
inferences can transmit knowledge will involve us in a host of difficult issues regarding the
Closure of knowledge under known entailment.
16
Fortunately, since the aim of this paper is only
to deal with basic knowledge, we can set these problems to the side. Yet a proper account of
Closure (whatever this amounts to) will spell out the conditions under which knowledge of a
proposition can be gained on the basis of knowledge of other propositions. Thus, a correct
account of Closure will give conditions for inferential knowledge that will include a guard
against accidentally-true beliefs that are generated by inferential reasoning. So inferential
knowledge will not escape the need for a non-accidentality clause – that clause will just be part

15
Setiya raised this concern in personal conversation.
16
Recall: Closure Principles attempt to specify the conditions under which one can have knowledge that Q when
one already has knowledge that P, and P entails Q.
78

of the Closure principle.
17
Thus, we have no reason to think that inferential knowledge will be
easier to come by than basic knowledge.
We will now see why, if BNAD is true, basic knowledge of non-natural moral properties
is impossible.

5. Why Beliefs in Non-Natural Properties are at Best Accidentally True
Non-natural moral properties are those that are not causally efficacious, and are,
metaphysically, “over and above” the natural properties, and hence are neither constituted by nor
constitutive of natural properties. As a result, non-natural moral properties do not enter into
constitutive or causal relationships with any wholly natural states of affairs. This is a problem
because an agent's having a certain disposition to form beliefs is a wholly natural state of affairs
(since this disposition is constituted by a brain state, and brains are natural). Thus, an agent's
using a method of forming reliably true beliefs about moral properties is a wholly natural state of
affairs. But according to BNAD, basic knowledge requires the reliability of the belief-forming
method to directly explain why the method is used. So the moral non-naturalist must hold that
the reliability of our methods for forming beliefs about morality explains why we form moral
beliefs in the way that we do, even though the moral facts cannot enter into constitutive or causal
relations with our methods for forming moral beliefs. As I will now argue, that is impossible.
To show that it is impossible for the non-naturalist to satisfy BNAD, let's first divide up
the logical space of ways in which BNAD can be satisfied. We first ask the question of what
direction the explanatory chain runs in: does the reliability of the method explain the use of the

17
It is common for Closure Principles to say that the subject must have competently deduced the conclusion from
the premises, and it is plausible that competent deduction consists in not only applying the correct inferential
processes, but applying them for the right reasons as well.
79

method, or vice versa? If the reliability of the method explains the use of the method, we can ask
two further questions. First: Is the type of explanation here causal, constitutive, or neither?
Second: Does 'the reliability of M' refer to the (statistical) reliability of M, as such, or are we
referring to the set of moral truths whose correspondence with the beliefs generated by M
constitutes the reliability of M? (That is: does the reliability that we are interested consist in the
higher-level pattern of belief corresponding to fact, or does it consist in the lower-level,
individual facts that ground this higher-level pattern?) Asking these questions allows us to
subdivide the logical space of ways of satisfying BNAD into six possibilities:

All six possibilities, (A)-(F), are unappealing. (A) says that the fact that S uses M
explains the reliability of M in some way. If use explains reliability (as in the rare case where the
(A)
(F)
(B) (D) Facts
(C) (E) Statistics
Other Causal Constitutive
1: Does
the
reliability
of the
method M
explain the
use of M?
Or does
the use of
M explain
the
reliability
of M?
Use
Explains
Reliability
Reliability
Explains
Use
3: Is it the
statistical
reliability that
does the
explaining?
Or the facts
themselves?
2: Is the kind of explanation relevant here causal,
constitutive, or something else?
80

facts are constituted by a subject or subjects' methods for belief-formation), that would satisfy
BNAD.
18
However, such an account isn't the kind of thing that a moral realist should adopt. If
the reliability of M depends on S's use of M, that would make the moral facts dependent on our
methods of forming beliefs about the moral facts. This doesn't just mean that the moral facts
would depend on our mental states; it means that the moral facts would depend on our methods
for forming beliefs about moral facts. So this approach would sacrifice the mind-independence of
morality.

Hence, moral realists should reject (A).
Since (B) involves moral facts entering into causal relationships with natural states of
affairs, and (D) involves the moral facts being constitutive of natural states of affairs, (B) and (D)
are ruled out by the nature of non-naturalism. (C) makes a category mistake: statistical reliability
isn't the kind of thing that can have causal powers, only the individual events can serve as causes.
(E) is the interesting possibility. We'll discuss it in the next section.
The sixth and final possibility, (F), is that the reliability of M could explain S's use of M,
where that kind of explanation is neither causal nor constitutive. I mention this final possibility
only in order to complete the logical space; no other kind of explanation seems promising.
If (A)-(F) are all untenable for the moral non-naturalist, then it is impossible for any of
our basic moral beliefs to be Well-Explained. Since Well-Explained belief is a necessary
condition on knowledge, therefore, basic knowledge of non-natural moral properties is
impossible. This, I believe, is the central epistemic problem for the moral non-naturalist.
19

18
This is, perhaps, how we are reliable in forming beliefs about our own mental states. And some constructivists –
like Street, perhaps – will find this a useful account of how our moral beliefs could be reliably and non-
accidentally related to the moral facts.
19
Replacing every instance of the word “moral” in the preceding paragraph with the word “mathematical” should
result in a sound argument as well, and I also believe that the resulting argument would be a good way to
understand the famous Benaceraff problem for non-naturalist mathematical epistemology (cf. Benaceraff 1973).
Accordingly, the Benaceraff problem makes moral and mathematical realists companions in guilt, rather than
81

6. Constitutive Reliability
Of the six possible ways to satisfy BNAD, the fifth possibility, (E), is the most promising.
(E) is the view that it is constitutive of our use of a particular belief-forming method that that
method be reliable. Let's call this view Constitutive Reliability. While no philosopher has clearly
and explicitly advanced this kind of view regarding moral knowledge, Wedgwood (2007)
defends a closely related view about standards of rationality (including, particularly, rationality
in moral belief) which would entail Constitutive Reliability with only modest and plausible
bridge principles. Consequently, I will spell out the commitments of Constitutive Reliability in
accordance with Wedgwood's views on the individuation of mental states and the relationship
between intentionality and normativity.
20

The Constitutive Reliability view is motivated by a general view regarding how our
mental states are individuated. According to this view, our intentional mental states are
individuated (in part) by the conditions to which they arise in response. So, for instance, part of
what it takes to feel fear is that one typically manifest this state in response to the right kinds of
things, i.e. things that are genuinely fearful. As the name “Constitutive Reliability” suggests, this
disposition to be reliable in one's fear judgments is constitutive of what it takes to be in a state of
fear. So any agent who is not reliable in their fear judgments is not really feeling fear, even if
they experience a surge of adrenaline, break out into a cold sweat, manifest a (defeasible)

companions in innocence as is sometimes assumed (cf. Clark-Doane 2014).
20
Wedgwood is not the only philosopher to have entertained the idea that some mental states might be
constitutively reliable. Sydney Shoemaker (1963) defends a version of Constitutive Reliability about the mental
state of remembering, saying that it is constitutive of this mental state that it be reliably formed. And
Wedgwood's view is motivated by the idea from Tyler Burge (2003) that perceptual states are individuated by
their correctness conditions. But to my knowledge, Wedgwood is the only philosopher who has attempted to
apply this idea in the moral domain.
82

disposition to shriek and run away, believe that some particular object is dangerous, and have
that particular uneasy feeling that we associate with fear.
21
They would instead be manifesting a
completely different state that happens to be similar to fear in a large number of ways. Call this
other mental state fear*.
As this account of individuation conditions for intentional mental states can be generally
applied, it can also apply to the mental state of having a moral intuition. For a defender of
Constitutive Reliability about moral intuitions, this mental state is also individuated by its
correctness conditions, and the correctness condition for a moral intuition (or any other mental
state that causes us to form a belief, for that matter) is that it be disposed to arise in response to
cases where the corresponding belief is likely to be true. So just as it is constitutive of the mental
state fear that it typically arise in response to situations that are genuinely fearful, it will be
constitutive of the mental state having an intuition that such-and-such is wrong that such-and-
such typically is wrong. And if our method for forming our moral beliefs is reliance on our
intuitions (which, in many cases, it seems to be), then it will be constitutive of this method that it
is reliable. So Constitutive Reliability satisfies BNAD.
However, Constitutive Reliability faces troubles on two other fronts. The first problem is
that it has a problem with higher-order knowledge. This is because the traditional hallmarks of a
mental state – its phenomenology and its downstream causal effects – do not suffice to
individuate that mental state, since the upstream antecedents of the mental state are also relevant.
Hence, according to Constitutive Reliability, there will be (at least) two distinct mental states
with the same phenomenology and downstream causal effects, that are differentiated only by the

21
Note: I am not taking any stand on which of these elements is constitutive of fear. This list is intended only to
draw a distinction between the Constitutive Reliability view and other views regarding the individuation
conditions for mental states.
83

situations to which they arise in response. So fear and fear* will be different mental states, and
so, too will be having a moral intuition and having a moral intuition*, where moral intuitions*
are mental states with the exact same phenomenological and downstream effects as an intuition,
but which are not reliable indicators of moral truth. In fact, they may well be anti-reliable
indicators of moral truth. But, because intuitions and intuitions*, ex hypothesi, differ only in their
reliability, they are subjectively indistinguishable; what it's like to have an intuition is exactly the
same as what it's like to have an intuition*. Thus, one will never be in a position to know
whether one is having a moral intuition or a moral intuition*. And because, according to
Constitutive Reliability, one has moral knowledge only if one's moral beliefs are the product of
moral intuitions (as opposed to intuitions*), one will never be in a position to know that one
knows. Of course, one might be able to figure out whether their mental state is an intuition or an
intuition* by having some independent confirmation of when they are getting things right and
when they are getting things wrong. But if reliance on intuitions (as opposed to intuitions*) is
our only route to basic moral knowledge, independent confirmation will be impossible.
Even if Constitutive Reliability gives an adequate story of how we know moral facts, the
structure of the view precludes us ever knowing that we know moral facts. But it is highly
counter-intuitive to think of first-order knowledge as rather easy to come by, while second-order
knowledge is impossible. To be clear: particular failures of second-order knowledge should not
be particularly troubling and are perhaps inevitable. But the systematic impossibility of second-
order moral knowledge is more worrisome. Second-order knowledge seems like the kind of thing
that should, generally, be possible. If everyone who believes that killing is wrong on the basis of
a moral intuition to that effect knows that killing is wrong, surely someone should be able to
84

know that they know this.
The second problem for Constitutive Reliability is that it is just not particularly plausible
that the mental state of having a moral intuition is individuated in the way that the view says.
Consider the case again of someone with fear* - someone who experiences all of the hallmarks
of fear (phenomenology, etc.) in response to brisk autumn days and fluorescent lights, but not
axe murderers. It seems clear that what the person in question is feeling really is fear, and not
some facsimile; she is just afraid of some silly things.
Similarly, consider someone who feels a certain sense of outrage at acts of particular
kinds, and who takes this sense of outrage to be grounds for censuring the person performing
those acts, or for condemning acts of this kind, and who will in turn form moral beliefs on the
basis of this sense of outrage. Anyone who feels this sense of outrage, with the attendant
dispositions to censure, condemn, and believe “that is wrong” (etc.) – should count as having a
moral intuition. And this will be true regardless of whether or not the agent having this intuition
is reliably disposed to form true moral beliefs. It is simply not plausible that whether or not this
feeling even counts as a moral intuition is a function of whether or not this intuition is reliable.
Thus, it is not constitutive of having a moral intuition that this intuition be reliable. So
while Constitutive Reliability has the structural features to solve the non-accidentality problem
for moral knowledge, it does so only at the cost of generating implausible conclusions about the
individuation conditions of the mental state of having a moral intuition. Accordingly, the view
should be rejected as inadequate on other grounds.
Since Constitutive Reliability was the only serious contender for how the moral non-
naturalist might satisfy BNAD, the fact that this view is false means that BNAD cannot be
85

satisfied by the moral non-naturalist. It follows that basic moral beliefs can only be accidentally
true, and, therefore, that basic knowledge of non-natural moral properties is impossible.
This does not mean that all moral knowledge is impossible for the non-naturalist, as the
possibility remains open that we could have inferential knowledge of non-natural moral
properties. However, since basic moral knowledge is impossible, any inferential moral
knowledge would have to be ultimately grounded in basic non-moral knowledge. Thus, moral
knowledge will only be possible if there is a mode of inference that can cross the infamous 'is'-
'ought' gap. We will return to this question in Chapter 5.

86

Chapter 4: What Makes Evolution a Defeater?
Among evolutionary psychologists, it is a popular hypothesis that human moral beliefs
have been shaped by evolutionary pressures. Populations that practice some form of reciprocal
altruism and engage in moralized patterns of thought can be shown through computer modeling
to have a distinct survival advantage over those populations that do not. If this hypothesis were
true, what would follow? It has become increasingly common in recent years to worry that the
truth of this evolutionary hypothesis would serve to “debunk” our moral beliefs. If moral beliefs
are the product of evolution, that could serve to show that our moral beliefs are not justified, that
they are not true, or that they do not amount to knowledge. But it's not entirely clear why this
result would follow. What is it, exactly, about the evolutionary origins of moral beliefs that
would create problems for realist views in metaethics?
In this chapter, I will argue that knowledge of the evolutionary origins of morality serves
to undermine our justification in beliefs in non-natural moral properties. Evolution, I argue, is a
defeater. My argument proceeds in three parts. In the first part, I examine the recent debate over
evolutionary debunking arguments and identify a number of flaws that are present in existing
debunking arguments. This discussion reveals three criteria that any debunking argument must
meet in order to be successful. In the second part (Sections 2 and 3), I look at the very general
question of what it takes for something to be a defeater, and develop a fully general account of
undercutting defeat. And in the third part, I argue that this account of undercutting defeat proves
that evolution is a defeater (Section 4), and does so in a way that meets the three criteria of
successful debunking arguments (Section 5).

87

1. Existing Debunking Arguments, the Good and the Bad.
Evolutionary debunking arguments come in two general flavors. Ontological debunking
arguments proceed from the premise that our moral beliefs are the product of evolution and
conclude that moral facts do not exist. These arguments generally proceed by appeal to Ockham's
Razor. Because we can explain many features of our moral belief and action by appeal to purely
natural facts about survival and genetic drift, moral facts serve no explanatory role, and thus can
be excised from our ontology. Michael Ruse (1986), Richard Joyce (2006), and Jonas Olson
(2014) all give some version of this argument.
1
Epistemological debunking arguments, on the
other hand, argue that evolution shows that our moral beliefs are not justified, or do not count as
knowledge. The locus classicus for concerns of this kind is Sharon Street's “Darwinian
Dilemma.” Street’s central dilemma is this: either our moral beliefs track the moral facts, or they
do not. The first horn of the dilemma is empirically falsifiable, says Street. Moral beliefs do not
track the moral facts; they track evolutionary facts. This puts us on the second horn of the
dilemma, where our moral beliefs do not track the moral facts. But, given that our moral beliefs
do not track the moral facts, it would be a remarkable coincidence if our moral beliefs were
mostly true. If we are not to believe in remarkable coincidences of this kind, we should conclude
that our moral beliefs are unlikely to be all or even mostly true. Accordingly, we are not justified

1
Although Richard Joyce seems to be making an ontological debunking argument in his The Evolution of
Morality, he has recently claimed (forthcoming) that his concern is, and always has been, epistemological. We
should, of course, take him at his word. Despite this, it is not clear how to read Joyce as an epistemological
debunker. He offers a number of interesting analogies and metaphors, but is highly inconsistent in what he takes
to be the core debunking principle. The most consistent throughline in Joyce’s work is that the anti-skeptic must
beg the question against the skeptic in order to answer debunking concerns. But as we saw in Chapter 1, whether
or not the anti-skeptic begs the question isn’t really what’s at issue (see also Berker (forthcoming)). What I will
argue in this paper is similar to some things that Joyce has said, but it is also very different from a number of
other things Joyce has said. I do not consider Joyce's debunking arguments in the main text because it is hard to
extract a core principle from them, and engaging in exegesis of Joyce will take us too far afield. But to the extent
that he is trying to articulate the same ideas as Street and Bedke, who I do explicitly discuss, we are all on the
same side.
88

in holding any moral beliefs. Importantly for Street, this skeptical conclusion can only be
reached if moral realism (which Street defines as the view that moral facts are mind-
independent) is true. If moral facts are not independent of us or our attitudes, then it would not be
a coincidence if our moral beliefs were largely true, as the moral facts would track our moral
beliefs (rather than the other way around). Street concludes that the only options views that hold
that moral beliefs are mind-dependent, or skepticism; Street opts for mind-dependence.
As I have been doing throughout the dissertation, I will set moral anti-realism and moral
naturalism aside and focus on the question of whether evolution makes epistemic problems for
non-naturalist moral realism. In this way, my thesis is closer to the one advanced by Matt Bedke
in his “Intuitive Non-Naturalism Meets Cosmic Coincidence.” The problem for non-naturalist
moral realism, Bedke argues, is that it positions moral facts outside the causal order. Thus, if we
hold the causal order fixed, our moral beliefs will be held fixed (since what beliefs we have is a
fact that exists within the causal order), but the moral facts will not remain fixed; they can freely
vary. Given that the moral facts vary independently of our moral beliefs, it would be a “cosmic
coincidence” if our moral beliefs are even roughly on the right track. The familiar skeptical
conclusion follows.
Street and Bedke draw our attention to the fact that our moral beliefs are the product of
naturalistic processes, and many have found this observation to be disturbing. But their
arguments have substantial weaknesses. Everything depends on the epistemic principle that the
evolutionary debunker defends – this principle must be true and it must yield the result that the
evolutionary etiology of our moral beliefs really does have skeptical implications. Unfortunately,
none of the suggestions made by Street or Bedke have these features.
89

In both Street and Bedke's work, the concept of a “coincidence” features prominently. For
Street, it would be a “remarkable coincidence,” if our moral beliefs are even roughly on the right
track, given the way our moral beliefs track evolutionary facts. And for Bedke, the fact that non-
natural facts exist outside the causal order means that it would be a “cosmic coincidence” if our
moral beliefs tracked the moral facts. Both Street and Bedke seem to assume a principle like the
following:
Coincidence: If it is just a coincidence that S's belief that p is true, and S knows this, S is
not justified in believing that p.
Ultimately, I think that neither Street nor Bedke are committed to the truth of Coincidence.
2
And
it is good that they do not endorse Coincidence, because Coincidence is false. Consider the
following counter-example from Kieran Setiya:
DMV: Charlotte goes to the DMV to get her driver's license renewed. Upon entering the
DMV , she looks around and is shocked to see that every member of her department is
also at the DMV . Charlotte knows immediately that this is a shocking coincidence. But
she blinks, rubs her eyes, looks twice, and confirms – every member of her department
actually is at the DMV .
Charlotte is justified in believing that every member of her department is at the DMV , even
though it is a coincidence that this is true, and Charlotte knows this. So Coincidence is false.
A further problem for a debunking argument based on Coincidence is that, contra Street
and Bedke, it's plausibly not a coincidence that our moral beliefs are true. A group of responses
to evolutionary debunking arguments, collectively known as Third Factor Accounts, hold that

2
Nonetheless, they have been read as endorsing this principle by several critics. See, e.g., Wielenberg (2010),
White (2010), or Setiya (2013, Ch. 2).
90

there it is no coincidence that our moral beliefs are correct because, while our moral beliefs
might not directly track the moral facts, they indirectly track those facts. As David Enoch has
argued, our moral beliefs about what is good track facts about what promotes survival – that's an
essential part of the evolutionary story about the origins of our moral beliefs. But, says Enoch,
survival is good. So there is a third factor – survival – that explains both our moral beliefs and
the moral facts. Our moral beliefs are thereby linked to the moral facts through this common
explanation. So it is no coincidence that our moral beliefs are largely true. Different varieties of
Third Factor Account identify different common explanations, but the details don't matter all that
much. What matters is the availability of a common explanation of both moral belief and moral
fact that would ensure a reliable connection between the two.
If Street and Bedke do not endorse Coincidence, then what do they endorse? In her
“Darwinian Dilemma” paper, Street emphasizes the way that evolutionary explanations of belief
reveal a troubling coincidence. Yet because Coincidence is false, in later work, Street advances a
different epistemic principle. The problem, says Street, is that moral realists cannot provide us
with a reason why our moral beliefs are reliable. This demand for a reason to believe in the
reliability suggests:
Explanation: If S's belief that P is to be justified, then there must be an
explanation of the fact that S's belief that P was reliably formed.
Explanation is a better epistemic principle than Coincidence, because it does not generate
the wrong results in the DMV case. We can give an explanation of why Charlotte's belief that
every member of her department is at the DMV is reliably formed: Charlotte formed this belief
on the basis of visual perception, which is a reliable belief-forming process. But Explanation
91

does not seem like a strong enough principle to support the debunking conclusion. As Enoch and
other advocates of Third Factor Accounts point out, the fact that survival is good can provide an
explanation of why our moral beliefs are reliably formed.
Street anticipates this objection, calling these kinds of explanations “trivially question-
begging” (Street ms, p. 17). And there does seem to be something to this charge – Enoch is
supposing that survival is good, which is a substantive first-order moral belief. But evolutionary
debunking arguments are supposed to be challenging our substantive first-order moral beliefs, so
appeal to those very beliefs seems a lot like cheating. Street seems to have in mind something
like the following:
Non-question-begging-Explanation: If S's belief that P is to be justified, then (1) there
must be an explanation of the fact that S's belief that P was reliably formed and (2) that
explanation cannot include the fact that P.
The problem here is that Non-question-begging-Explanation is also a false principle. As
Selim Berker (2014) and Katya Vavova (2014a) have pointed out, it is always impossible to
explain how any faculty is reliable without making some assumptions about the truth of beliefs
delivered by that faculty. One cannot explain how our beliefs about the external world are
reliable without making some assumptions about the way the external world is. (Consider: if we
cannot assume that any of our beliefs about the external world are true, then we would have no
grounds for thinking our sense perceptions are reliable). So in attempting to articulate an
epistemic principle that is strong enough to entail moral skepticism, Street has advanced an
epistemic principle that entails global skepticism. That’s a problem.
Given the failure of Street's arguments (as given), we might think that Bedke's argument
92

provides a better template for formulating a debunking argument. Unfortunately, Bedke's
argument also suffers from a fatal flaw. Bedke argues that what generates epistemic concerns via
“cosmic coincidence” is the fact that if the causal order – and nothing else – is held fixed, then
our moral beliefs will be fixed, but the moral facts will not. Thus, for Bedke, knowledge is
destroyed whenever the belief in question is held fixed a result of causal processes, but the fact in
question is not.
I emphasize that it is the causal order and nothing else that must be held fixed because
Bedke's argument depends on it. If hold the causal order and any metaphysically necessary facts
fixed, for instance, then the moral facts will be fixed as much as our beliefs, since the moral facts
strongly supervene on facts that are part of the causal order. This observation draws attention to
how artificial it is that Bedke is holding fixed only the causal order. Bedke's argument guarantees
that all and only facts that exist outside of the causal order will be defeated by cosmic
coincidence. But Bedke gives no independent motivation for thinking that the holding-the-
causal-order-fixed test will generate any epistemically interesting results, or any reason for
thinking it will generate more relevant results than a holding-the-causal-order-and-some-other-
things-fixed test. Bedke's argument relies on an epistemic principle that is ad hoc, and is for that
reason unacceptable.
So while the evolutionary history of our moral beliefs might seem initially troubling,
existing arguments do not advance an epistemic principle that is true, well-motivated, and yields
a skeptical conclusion about morality in light of the possibility of offering a Third Factor
Account. In the rest of this chapter, I will attempt to offer an argument that lacks these flaws.

93

2. The Nature of Undercutting Defeat
There are several ways in which one could argue that the evolutionary history creates
epistemic problems for our moral beliefs. In this chapter, I will argue that this evolutionary
history is an undercutting defeater for our moral beliefs. While we may be prima facie justified
in having certain moral beliefs on the grounds that those beliefs seem pre-theoretically obvious,
or are justified by strong moral intuitions, or something similar, once we learn about the
evolutionary history of our moral beliefs, we are no longer justified in believing them.
Accordingly, we need an account of what undercutting defeaters are, so that we can have a
neutral framework for figuring out whether or not evolution would constitute a defeater of this
kind.
The notion of an undercutting defeater is a common one in epistemology, but it is rather
difficult to give a precise explication of what, exactly, an undercutting defeater is, as opposed to
a rebutting defeater. Rebutting defeaters and undercutting defeaters are supposed to be the only
kinds of defeaters that there are. The basic idea behind the distinction is that, for some piece of
evidence E in favor of a proposition P, rebutting defeaters are best thought of as attacks on P,
while undercutting defeaters are attacks on E (or, more precisely, the support that E lends to P).
Rebutting defeaters, therefore, serve as evidence in favor of ~P, regardless of whatever other
evidence the subject might have. Undercutting defeaters, on the other hand, call into question the
evidential support relations between E and P. Undercutting defeaters show that E is not as good
evidence for P as we might have previously thought. Thus, according to Thurow (2013), “d is an
undercutting defeater for [P] iff d is a defeater for [P] which is (or is an epistemically appropriate
basis for) the belief that one’s actual ground or reason for [P] is not indicative of [P]’s truth” (p.
94

1592). And according to Sinnott-Armstrong (2006), “one reason undermines another when the
former does not provide any reason to believe anything to the contrary but still does make the
conflicting reason inadequate for justified belief” (p. 68).
These kinds of accounts of undercutting defeaters are good starting points, but they are
incomplete. Accounts like Sinnott-Armstrong's attempt to explain undercutting defeaters in terms
of what they do – they weaken intuitive evidential support relations – but they do not provide a
constructive account of what undercutting defeaters are. What relation must a piece of evidence
E, a proposition P, and an undercutting defeater D stand in such that D will prove E to be poor
support for P? Thurow's idea, that D implies that E is “not indicative of [P]'s truth,” is on the
right track, surely, but this idea needs substantially more refinement. What does it take for one
proposition to be “indicative of the truth” of another proposition?
For one concrete suggestion of how this notion can be spelled out, we look to Pollock
(1987). Pollock was the first to draw the distinction between undercutting and rebutting
defeaters. For Pollock, “[D] is an undercutting defeater for [E] as a prima facie reason for S to
believe [P] if and only if [D] is a defeater and [D] is a reason for denying that [E] wouldn't be
true unless [P] were true” (p. 485). So for Pollock, defeat arises whenever we have (new)
evidence that our (old) evidence does not counterfactually depend on the proposition that it is
supposed to support.
The trouble with counterfactual accounts of this kind is that they deal poorly with
necessary truths. If S's belief is a necessary truth, then it will be trivially true for all evidence E,
that E would not be true unless P were true. Thus, the only grounds for denying this
counterfactual must be grounds for denying either P's truth or its necessity. If one knows that the
95

truth one is searching for is a necessary truth (as when scientists attempt to find out what
constitutes a certain kind of substance), then the only evidence that can count against this
counterfactual would be evidence that P is not true. But since evidence that P is not true would be
a rebutting defeater for one's evidence that P, it follows that there will be no undercutting
defeaters for necessary truths.
3
And that is absurd. For illustration, consider the following case:
Mischievous Lab Assistant: A chemist is running an experiment to determine
the nature of fire. A pile of experimental data confirms – fire is the release of
phlogiston! But then the chemist learns that his lab assistant played a practical
joke on him: all of the data that was supposed to confirm the hypothesis that
fire is phlogiston was actually falsified by the lab assistant.
At this point, the chemist would no longer be justified in his belief that fire is phlogiston.
He'd need to rerun his experiments (and fire his lab assistant). A counterfactual account may be a
nice heuristic for when an undercutting defeater is present, but it can't amount to an exact
account of undercutting defeat. This result is particularly important when we consider the fact
that many moral truths (which are the object of our ultimate concern) are, purportedly, necessary
truths.
Another promising suggestion is to think of the relation between defeater, evidence, and
belief as probabilistic. Thus, roughly, E supports P just in case Pr(P|E) > Pr(P), and D is an
undercutting defeater for this supports just in case Pr(P|E) > Pr(P|E&D) >= Pr(P).
4
The problem
with this suggestion comes when we try to explicate the nature of the probability function

3
Note that this objection does not assume epistemic externalism. If S knows that P will be necessary if true, then
S's internally-accessible evidence guarantees that there will be no undercutting defeater for S's belief that P that
is not also a rebutting defeater.
4
See Kotzen (ms) for a sophisticated suggestion along these lines.
96

present in these formulations. If the notion of probability that is relevant here is supposed to be
some sort of objective probability, then the problem of defeating necessary truths will arise again
in full force. If P is necessary, Pr(P) = 1. It follows that nothing will be evidence for P and, more
damningly, no evidence in favor of a necessary truth could ever be undercut.
The natural response to this concern is to say that the probability function isn't supposed
to pick out any objective notion of probability, but instead be a normal Bayesian credence
function; people often have subjective credences in necessary truths that are less than 1. But now
a further question arises: what version of Bayesianism we are to adopt? We could be pure
subjective Bayesians, who hold that the only rational constraints on our credences are given by
Bayes' Theorem and the other axioms of the probability calculus. But if our credences are only
rationally constrained in this way, then prior credences (both conditional and not) are under no
rational constraints, so it's possible for one person's credence function to yield the result that D is
an undercutting defeater while another person's credence function does not yield that result, and
both agents will count as equally rational. In essence, a subjective Bayesian account will specify
which propositions different subjects take to be defeaters, but not which propositions actually are
defeaters. This is not to say that there will be no cases in which the question of what we take to
be defeaters will be relevant. But we should also acknowledge that there is another important
sense in which some propositions really are defeaters, no matter what attitude the subject takes
toward those propositions. If the chemist learns that his experimental data was falsified, this
creates important epistemic problems. If the chemist does not lower his credence in his
conclusions upon learning this (because his prior credences dictate that he should keep his
credence the same even after learning about this falsification), then the chemist's prior credences
97

are missing something important.
Most Bayesians are not purely subjective Bayesians, but accept some additional
“objective” constraints on what one's prior credences have to be. Either these additional
constraints constrain prior probabilities enough that the same propositions will come out as
defeaters for the same evidence for all agents, or they do not. If they do not, then the problems
for pure subjective Bayesianism still apply in full force. If, on the other hand, these additional
rational constraints suffice to determine what does and does not count as a defeater, we now need
an account of what facts ground these rational constraints on prior credences. What must the
relationship between D, E, and P be such that D defeats the evidence E provides for P? – but this
is just what we were looking for in the first place. This isn't to say that an objective Bayesian
framework is wrong, of course. It might be useful and, if the details are supplied correctly,
correct. But the viability of this framework will depend entirely on our ability to independently
characterize undercutting defeaters. So we keep looking.
Pollock and Gillies (2000) improve on Pollock's original, counterfactual account by
dropping reference to the counterfactual. According to Pollock and Gillies, “[u]ndercutting
defeaters attack the connection between the premises and the conclusion of a defeasible
inference rather than attacking the conclusion itself” (p. 75, italics in original).
5
The idea here is
that E supports P just in case E and P are connected in the right kind of way, and D would thus
count as a defeater if D were evidence that E and P are not connected in this way. But connected
how?
Fortunately, common language provides a useful suggestion about what the nature of this

5
Unfortunately, Pollock and Gillies go on to gloss this account in counterfactual terms. Fortunately, their initial
suggestion of a connection is one that will prove fruitful.
98

connection might be. The idea of an “undercutting defeater” is a bit of technical terminology
within philosophy, and as such it is not the phrase that typical speakers of English use to describe
the phenomenon of undercutting defeat. Instead, we are more likely to talk about evidence as
being “explained away.” We use this notion of “explaining away” to talk about more than just
relations between evidence and belief. In general, something is said to be “explained away”
whenever we can give an explanation of a thing that mitigates or eliminates its importance. And
when evidence is “explained away,” what we are doing is giving an explanation of that evidence
that mitigates or eliminates its evidential importance.
We can put these two thoughts together, then, and say that E supports P just in case the
fact that P explains S's evidence, E. (Note: in normal circumstances, the existence of this kind of
explanatory connection will ground Pollock's counterfactual, “E wouldn't be true unless P were
true.”) And this, then, allows us to give an account of undercutting defeat. If E supports P just in
case P explains E, D will be an undercutting defeater for P just in case D is evidence that P does
not explain E – something else does. That is, D is an undercutting defeater just in case it is
evidence for some rival explanation of E. This principle is in need of some refinement, but at a
first stab:
Explaining Away Defeats (EAD): Evidence in favor of an explanation of
some set of S’s evidence E in favor of a proposition P will be an undercutting
defeater for S's belief that P just in case the explanation of E does not involve
P.
EAD can explain the instance of undercutting defeat in Mischievous Lab Assistant, and it
does so in a quite natural and straightforward way. The reason that the chemist learning of his lab
99

assistant's indiscretion is a defeater is because the chemist has learned that the explanation of his
data does not include the fact that fire is phlogiston. That is, he does not have the evidence that
he has because fire is phlogiston. Rather, he has it because his lab assistant wanted to play a
trick.

3. Refining EAD
In this section, we'll look at a few challenges to EAD as an account of undercutting
defeat. While I will motivate EAD (or the version of EAD that emerges from this section) as
giving necessary and sufficient conditions for undermining defeat, I only intend to defend the
sufficiency of EAD for specifying circumstances when something counts as an undermining
defeater. This is because I am more convinced of the sufficiency of EAD than its necessity, and
because my Epistemological Debunking Argument only needs to specify a sufficient condition
for undercutting defeat. Accordingly, although some of what I say will serve to motivate the
necessity of EAD, in this section I will try to show that EAD does not overgenerate defeat.
3.1 Knowledge of the Future
A first concern one might have is that EAD rules out knowledge of the future. I know
many things about the future, and, accordingly, I have evidence for many propositions about the
future. But none of my evidence about the future is explained by these future facts. The future
facts certainly don't cause me to have any of my evidence and it doesn't seem as though the
future facts explain my present beliefs in any other way. My evidence regarding future facts is
contained entirely in the present and the past. For example, I know that the sun will rise
tomorrow on the basis of the fact that it rose today, and yesterday, and the day before that, and so
100

on. Thus, EAD seems to rule out knowledge of the future, since all my evidence for my beliefs
about the future is not explained by those future facts, and I know this.
But when we consider this sunrise case, a very natural response to the objection presents
itself. The explanation of all of my evidence that the sun will rise tomorrow is just this: that the
sun has a remarkable tendency to rise every morning (or at least that it has had this remarkable
tendency up until today). But from this explanation of my evidence, one may inductively infer
that the sun will rise tomorrow. In what sense, then, is this a rival explanation of the evidence? It
may not include P (that the sun will rise tomorrow) as a constituent, but since this explanation of
the evidence can justify an inference to the fact that P, learning about this explanation should not
make me less confident that P is true – it should make me more confident that P is true, since I
can infer it from my new information!
Accordingly, when EAD tells us that we have a defeater when we learn that the
explanation of our evidence does not “involve” the proposition that P, we should understand this
to mean that an explanation “involves” the fact that P just in case P can be inferred from that
explanation, where “can be inferred” just means that there is a valid inference process that takes
us from the explanation of the evidence for P to the fact that P.
6

If P is a part of the explanation of the evidence, then the explanation deductively entails P.
Thus, all cases where P is part of the explanation of the evidence will be cases where P can be
inferred from the explanation. But because an explanation “involves P” when P can be inferred

6
Propositions are known inferentially just in case there is some prior knowledge among S's grounds for believing
that P. In this case, this prior knowledge would include S's knowledge of the explanation of E. This is a stronger
notion of inference than that deployed by Trompan (2011) or Huemer (2005), who hold that a proposition is
known inferentially if and only if one has grounds for believing it. This other conception of inferential
knowledge is far too weak to be useful: since one must have some grounds for believing that P in order to know
that P, it would follow from Huemer and Tropman's account of inferential knowledge that all knowledge is
inferential.
101

from the evidence by any mode of inference – including, for instance, inductive inference –
knowledge of the future is not always defeated.
Note, too, that EAD can still provide an explanation of how evidence for propositions
about the future can be undercut. If I learn that my evidence was taken from a non-representative
sample of days on which the sun has risen, or that somehow the laws of nature are bound to
change overnight, then the explanation of my evidence is not one from which P can be inferred;
one cannot infer that the sun will rise tomorrow from the fact that the sun has risen in a non-
representative sample of past days. So it looks like thinking of P-involving explanations as
explanations from which one may correctly infer P gets the right results in this case.
Since what matters is what S is in a position to know, the set Γ from which S “can infer”
P need not include only propositions that are part of the explanation of S's evidence for P. Γ can
include other background information that S already knows. But allowing background
information to be part of S’s evidence from which S can infer P suggests another objection.
Consider the following kind of reasoning: Vera sees a wall that looks red. She then, justifiably,
comes to believe that the wall is red. So her background information contains the following two
justified beliefs: The wall looks red, and the wall is red. Vera then learns that a red light is
shining on the wall. But, from the explanation of her evidence (there is a red light shining on the
wall) together with her justified background beliefs (the wall is red), she can infer that the wall is
red; this follows deductively from her justified background beliefs. So because she can infer that
the wall is red from the explanation of her evidence together with her background beliefs, this
explanation of her evidence is not a rival explanation. Hence, learning that a red light is shining
on the wall is not a defeater.
102

This is obviously a bad reasoning process, and it's clear where the mistake lies: it was
incorrect to include the proposition that the wall is red among the propositions that constitute an
acceptable ground for inference. To prevent this kind of circular reasoning, we should say that S
can infer P from the explanation of E just in case S can infer P from the union of the explanation
of E and S's independent background knowledge, where background knowledge is independent
just in case it is not E or P, nor is it known on the basis of either E or P.
Note that this “independence” qualification does not mean that one must always have an
independent reason to believe that P, when one believes that P. That is, this is not a requirement
that all beliefs must be independently confirmed – that requirement would lead us quickly into
global skepticism. What this “independence” qualification means is that, if one receives an
explanation of one’s evidence, E, that P, and that explanation does not contain P as a constituent,
then one may not appeal to either E or P in an attempt to infer P. This is the minimal change
needed to preserve the coherence of any account of undercutting defeat.
3.2 Restricted Explanations
A third concern one might have about EAD is that it is far too easy to give non-P-
involving explanations of my evidence, simply by cutting out certain parts of the explanation.
For instance, suppose Tom has a visual experience as of a cat on the mat. Tom believes that the
cat is on the mat, his visual experience serves as his evidence, his vision is functioning normally,
and he is not in abnormal circumstances. So we can explain Tom's evidence by appealing to the
fact that there is a cat on the mat. The fact explains the evidence, so no defeater is present, which
is as it should be. But here is another explanation that we can give of Tom's visual experience.
Light waves strike Tom's retina, in such a way that the photosensitive cells in Tom's retina are
103

stimulated, which in turn stimulates the optic nerve and the visual cortex of the brain. This is a
perfectly good explanation of Tom's visual experience, but it doesn't involve either cats or mats;
it's an explanation purely in terms of Tom's visual sensory system. But Tom learning how his
visual sensory system works should not defeat his belief that there is a cat on the mat.
It's easy to see what is going on here. Tom's belief is undefeated because the explanation
that we've given of his visual experience is only a partial explanation. It leaves out portions of
the proper explanation of his visual experience, and these portions contain the fact that the cat is
on the mat. In order to block counterexamples of this kind, then, we should say that it is only
evidence in favor of complete non-P-involving explanations that serves as a defeater, where an
explanation is complete just in case (1) the explanation is sufficient for the explanandum and (2)
the explanation stretches far enough back in time to cover both proximate and ultimate causes of
the explanandum. The information about Tom's visual sensory system satisfies (1), but it is not
complete because it does not satisfy (2). The explanation is not complete in that it does not tell us
why light struck Tom's retina in the way that it did. A complete and accurate explanation of
Tom's visual experience will include the fact that the cat is on the mat, and thus it will be a P-
involving explanation. Learning about that isn't a defeater.
Note that this does not mean that one must be in possession of a complete explanation of
one's evidence in order for that evidence to have justificatory force. Even in cases where a
subject has no complete explanation of her evidence available, she may still be justified. What
EAD says is that if a subject receives a complete explanation of her evidence, that explanation
had better be P-involving, or else the evidence in favor of this explanation will be an
undercutting defeater.
104

3.3 Cheap Evidence and the Strength of Defeat
Another concern that someone might have about EAD is the worry that evidence is
cheap. According to EAD, in order for there to be no undercutting defeater for one of my beliefs,
P, I must have no evidence in favor of any explanation of my evidence for P that does not involve
the fact that P. But this seems too stringent. I have a lot of evidence that can be called on to
support any number of propositions, and there are a lot of potential explanations of my evidence
that P that do not involve the fact that P. Given this, it would be surprising if none of my
evidence supported any non-P-involving explanation of my evidence. Yet EAD says that this
must be the case if my belief that P is to be undefeated.
The problem with this objection is that it conflates the existence of a defeater with the
existence of a strong defeater. Defeaters, as we've seen, are a kind of evidence, and evidence can
come in all strengths. Accordingly, defeaters themselves can come in all strengths.
7
Strong
defeaters are defeaters that have enough defeating force to render a belief unjustified. Weak
undercutting defeaters, on the other hand, serve only to mitigate the evidential force of that piece
of evidence. It is absurd to think that all of our evidence for all of our beliefs is strongly defeated.
But it is possible – indeed, I think, likely – that the vast majority of our beliefs admit of some
weak defeat.
Furthermore, there is a very natural way of spelling out when a defeater is strong or
weak, according to EAD. EAD says that evidence in favor of rival explanations of the evidence
that P is a defeater. So it is quite plausible to say that the strength of the defeater is directly
proportional to the strength of the evidence in favor of a rival explanation. If the chemist gets a
faint hint that his data might have been falsified, he ought, rationally, to be slightly less confident

7
As Schroeder (2011) argues, it is best to think of total defeat as a limiting case of evidential attenuation.
105

in his conclusions. If he learns conclusively that his data has been falsified, he ought, rationally,
to be much less confident in his conclusions. This is what our account of the weight of defeaters
suggests, and it seems like precisely the right thing to say.
Another way to think about this is that the strength of the defeater is a function of how
likely the defeater makes it that the actual explanation of E is not P-involving. Thinking of things
this way helps explain the import of complete explanations. A restricted explanation, of the
variety that we looked at in 3.2, won’t be the kind of thing that can rule out P being actually
involved in the explanation of E. But complete explanations provide much stronger support for
the conclusion that P is not involved anywhere down the line. Since the strength of the defeater is
proportional to how strong our evidence is that the explanation is not P-involving, the defeater
will be stronger the more complete the non-P-involving explanation is. Fully complete
explanations will be very strong defeaters. Highly restricted explanations will be very weak
defeaters or not defeaters at all (if they are restricted enough not to provide any evidence that the
explanation of the original evidence is not P-involving). Most partial explanations will fall
somewhere in the middle.
3.4 Conceptions of Evidence
In this chapter, I’ve referred repeatedly to S’s evidence, E, for some proposition P, and
asked under what conditions the support that E provides for P could be undercut. But what is this
evidence? I assume here only that evidence is that upon which justification depends, but there are
many views on justification, and how EAD should be interpreted depends on what view of
justification we take.
Three views are relevant here. The first, which we may call attitude-internalism, says that
106

evidence for some proposition consists in the beliefs or other mental states of a subject. It is the
beliefs themselves, not the contents of those beliefs, which count as evidence. So if I believe that
Sally was at the meeting on the grounds that I know that all faculty members were at the
meeting, my evidence that Sally was at the meeting is my belief that all faculty members were at
the meeting. A second view, that we may call content-internalism, says that in order for a
proposition to count as evidence, I must believe it, but it is not my belief that counts as evidence,
it is the content of that belief. So my evidence that Sally was at the meeting would be that all
faculty members were at the meeting (as I believe). Both of these views hold that evidence is a
function of a subject’s mental states, and that a subject cannot be justified in believing a
proposition unless that subject has some other mental states that back up that belief. The third
position to consider is externalism, which holds that some propositions are evidence not in virtue
of being the object of any mental state, but simply in virtue of being true. According to
externalist conceptions of evidence, the attitude of the believer toward the evidence is irrelevant.
So when EAD talks about a certain explanation of a subject’s evidence, what is being
discussed? Is it the explanation of a subject’s attitude toward a proposition, the explanation of a
proposition (as believed by some subject), or just the explanation of the truth of some
proposition? To this point, I have been assuming attitude-internalism (since this is the position
with which I am most sympathetic). So the mention of an “explanation of evidence” in EAD
should be understood as explanations of the mental states of the agents in question. If we accept
content-internalism, then we will need to modify EAD to reflect this different conception of
evidence, but the change that is needed is both obvious and unproblematic. If evidence is not a
belief in some proposition, but instead the proposition itself qua object of attitude of some agent,
107

then what counts as a defeater is not a rival explanation of the evidence, but a rival explanation
of the agent’s attitude toward that evidence. More colloquially, what is being explained on this
conception of evidence is not the evidence itself, but the fact that the agent has that evidence.
8

No matter what conception of evidence we have, defeat arises when we have a rival explanation
of the relevant mental state.
This means that externalism presents a challenge. If mental states are irrelevant to
justification, it’s hard to see how an S’s learning an explanation of some irrelevant mental state
could be a defeater.
9
But the issue is more general than that. If the truth of some proposition is
sufficient for S to be justified in believing that P, it’s hard to see how anything that S might learn
would count as a defeater for S’s belief. No matter what S learns, the justifying proposition is
still true,
10
and so S will still be justified. Thus, what first looked like a problem for EAD begins
to look like a problem for externalism. EAD cannot explain undercutting defeat if externalism is
true – but this is because, if externalism is true, defeat does not exist.
An externalist may be willing to bite the bullet here and say that, in fact, defeat does not
exist. (Indeed, Lasonen-Aarnio, who first noted this problem for externalism, takes this bullet-
biting approach).
11
So in the interests of remaining ecumenical, I will conditionalize my
conclusions here to say that EAD is a sufficient condition on undercutting defeat if anything is.
Putting this all together, we get EAD2.
EAD2: For any complete explanation A of a subject S's internally-accessible

8
See Feldman (1988) on having evidence.
9
Thanks to Jeff Russell for pressing me on this point.
10
Unless the justifying proposition is a proposition about S’s attitudes. But to say that the justifying proposition
for some subject is always a proposition about that subject’s attitudes is not externalism – it is attitude-internalism.
11
See Lasonen-Aarnio (2010) for more on externalism’s trouble with defeat and her corresponding argument
against the idea of defeat.
108

evidence E in favor of a proposition P, such that S may not infer P from A and
S's independent background knowledge, evidence in favor of A undercuts the
support that E provides to P.
3.4 Extensional Correctness
In the remainder of this section, I will provide additional support for EAD2 by bolstering
the case for EAD2's extensional correctness. First, I will show how EAD2 fails to generate a
defeater in normal cases of perceptual, memorial, conceptual, and inferential knowledge, which
is as it should be. Second, I will show how EAD2 successfully predicts defeat in a number of
very different cases where undercutting defeaters are present.
The fact that EAD2 does not generate defeat in cases of perceptual and memorial
knowledge has already been suggested, but it's worth going over the details. In typical cases of
perception, I have a perceptual experience as of P because my senses put me into causal contact
with the fact that P. The evidence for the belief is the perceptual experience. In normal cases, the
perceptual experience is explained by the fact perceived. So in normal cases, the explanation of
the evidence is P-involving. And in typical cases of memorial knowledge, I recall that P because
my memory has stored the fact that P, and my memory has stored the fact that P because P is true
(and I was earlier in a position to know this). So again, in typical cases, the explanation of my
memorial evidence that P will be P-involving. No defeater is present in either case.
Inferential knowledge is not much more complicated. In instances of inferential
knowledge, I know that Q, and correctly infer P on the basis of Q. Because I know Q, and
because P may be correctly inferred on the basis of Q, P may be inferred from the explanation of
my evidence. Thus, the explanation of my evidence for P will be P-involving, and so EAD2 will
109

not wrongly generate a defeater.
Finally, conceptual knowledge. In cases of conceptual knowledge that P, P is a conceptual
truth – that is, the truth of P follows from the concepts that are constitutive of the proposition P.
Accordingly, anyone who understands the concepts involved in P will be in a position to infer
that P. So anyone who understands P will be able to infer that P. And so, for a conceptually
competent individual, any explanation of anything will be a P-involving explanation. So EAD2
does not predict a defeater for genuine conceptual truths. (This is not to say that one's belief in a
genuine conceptual truth cannot be defeated in any way, only to say that EAD2 won't be what
accounts for any such defeater.)
I will now show that EAD2 does correctly generate defeat in a number of different cases
of undercutting defeat. In one classic example, the fact that something appears red is a reason for
me to think that that thing is red. But when I learn that the object in question is being illuminated
with red light, I am no longer justified in believing that it is red. This is because I have evidence
in favor of a non-P-involving explanation of my visual experience: an explanation in terms of a
red light, not a red object.
Another case, this time involving memory. Tiffany seems to remember that she went to
the movies for her birthday last year. But then Tiffany learns that she has been the victim of a
bizarre experiment a little over a month ago, which removed all of her memories and replaced
them with new ones. Tiffany now has evidence in favor of a non-P-involving explanation of her
beliefs about her birthday last year – a mad scientist explains her memories, not the fact that she
was actually at the movies. So EAD2 also explains why Tiffany's belief about the past is defeated
by learning about this experiment.
110

Another classic case, this time involving statistical reasoning. I know that Tweety is a
bird. This known proposition – that Tweety is a bird – is evidence for thinking that Tweety can
fly, and it is good evidence. This is because, given the correct background information about the
flight of birds, I am in a position to perform an inductive inference to the conclusion that Tweety
can fly. I then learn that Tweety is a penguin. I no longer have good reason to think that Tweety
can fly. When I learn that Tweety is a penguin, I have an explanation (that Tweety is a penguin)
of my evidence (that Tweety is a bird) from which I am not in a position to perform an inductive
inference to the conclusion that Tweety can fly. So I have evidence in favor of an explanation of
my evidence that is not P-involving. Once again, EAD2 correctly predicts a defeater.
Another case, this time involving testimony. Janet tells me that blue is her favorite color. I
then come to learn that Janet is an incorrigible liar, who says whatever suits her momentary
fancies regardless of the truth of her assertions. I now have evidence in favor of an explanation
of Janet's testimony that is not P-involving; Janet told me blue is her favorite color not because it
is her favorite color, but because it amused her to say that thing at the moment. EAD2 correctly
predicts a defeater.
Not only is EAD2 intuitive, as I argued in Section 2, it is also able to generate the correct
results regarding the presence of defeat in a wide range of cases involving different kinds of
knowledge and defeat. Since EAD2 generates accurate predictions both in cases where a defeater
is expected and one where it is not, there is a further case to be made in favor of EAD2 on the
basis of its explanatory power. I conclude that the case in favor of EAD2 is very strong.

111

4. The Darwinian Defeater
With EAD2 in hand, we are now in a position to see why evolution counts as a defeater
for our moral beliefs. I begin by assuming that evolutionary biologists have shown that forces of
natural selection suffice to explain why we have the moral concepts that we have, and why we
apply them in the way that we do. (In making this assumption, I am following Street in that I am
not attempting to adjudicate issues in evolutionary biology by myself, but am instead seeing
what would follow if a certain thesis in evolutionary biology is true). If this assumption is
correct, we can give an evolutionary explanation of all of our evidence in favor of any of our
moral beliefs. The fact that moral beliefs seem so intuitive, that most people can't bring
themselves to violate moral norms without significant distress, the fact that moral beliefs are so
widely shared throughout society – all of these can and have received Darwinian explanations.
And, of course, these Darwinian explanations are given entirely in natural terms. Evolutionary
explanations of our moral attitudes are given in terms of how these attitudes will be conducive to
survival (viz., of an individual who has these attitudes, living in a population of individuals that
share these attitudes), not in terms of rightness or wrongness, goodness or virtue. So where P is
any of our moral beliefs, evolutionary explanations of our evidence for our moral beliefs will be
non-P-involving. And there is a substantial body of evidence in favor of these evolutionary
explanations of our moral faculties. So by EAD2, that evidence serves to defeat our moral
beliefs. This is why evolution is a defeater.
More formally, the Evolutionary Debunking Argument from EAD2 goes like this:
(1) Complete Darwinian explanations can be given of why moral propositions
112

seem true to human agents.
12

(2) There is evidence in favor of these Darwinian explanations.
(3) One may not infer the truth of any moral proposition from these Darwinian
explanations and one's independent background knowledge.
(4) EAD2
(5) Therefore, human agents' beliefs in moral propositions are defeated by
Darwinian explanations.
One might object to Premise (1) on the grounds that Darwinian explanations are not
complete. A Darwinian story can supply the ultimate causes of our moral beliefs, but it does not,
and cannot, tell us anything about the proximate causes of those beliefs – the neuropsychological
mechanisms that are in play whenever a moral belief is formed. Furthermore, Darwin can at best
provide us with the “Nature” side of the explanation of why we are inclined to form the moral
beliefs that we are inclined to form, but it is widely agreed that both Nature and Nurture play
some role in determining many of our traits, including our dispositions to form moral beliefs.
This objection rests on a misunderstanding of the argument given here. As Street herself
has pointed out, there's nothing essentially Darwinian about the Darwinian Dilemma. The
importance of an evolutionary story is that it goes a long way toward completing a naturalistic
explanation of our moral beliefs, by providing the ultimate causes of moral belief to supplement
the proximate causes that are uncovered by sociology and neuropsychology. As we saw in 3.3
and 3.4, evidence in favor of a rival explanation is a stronger defeater the less room it leaves for
explanations of the evidence that are P-involving. Darwinian explanations rule out the possibility
that our culture and psychology are ultimately explained by a mind-independent, non-natural

12
I am willing to grant to the intuitionist that a proposition's seeming true can count as evidence that it is true.
113

moral truth.
I will not dedicate many words to a defense of either Premise (2) or Premise (3). The
truth or falsity of (2) is an empirical matter, and I am not in a position to make a case one way or
other on questions of evolutionary psychology. I will follow Street in saying that if the
evolutionary evidence is as I assume it to be, the argument goes through. For those interested in
the nature of this evidence, Richard Joyce (2006) has done impressive work in synthesizing a
mass of empirical evidence in favor of the Darwinian hypothesis. And Premise (3) will be the
subject of the next chapter. (3) is entailed by the proposition that inferential moral knowledge is
impossible, and that is the thesis of Chapter 5. For now, it is enough to note that (1), (2), and (4)
entail that justification of basic moral beliefs is impossible.
The last premise of the argument is EAD2. This premise was defended in Sections 2 and
3; I won't spend any more time on it here.
One upshot of this discussion is that the case that evolution is a defeater is much more
straightforward than one might have thought. We do not have to appeal to any dubious bridge
principles invoking notions of coincidence or question-begging reasons in order to show why
evolution is a defeater. The evolutionary defeater falls right out of our best account of what it
takes for something to be an undercutting defeater. The Darwinian defeater is a paradigm case of
undercutting defeat. If anything is an undercutting defeater, then Darwinian explanations of our
moral beliefs are undercutting defeaters. Accordingly, an opponent of evolutionary debunking
arguments must give an account of undercutting belief that correctly predicts undercutting defeat
in a wide variety of cases (as we saw EAD2 can do) without entailing that evolution is a defeater.
This is a hefty burden for those opposed to debunking arguments to take on.
114

EAD2 also helps explain the intuitive pull behind Street and Bedke’s arguments. Bedke
claimed that we should look at the variance of our moral beliefs while holding the causal order
constant. We can now see what was right about this thought – the “causal order” is important
because causal connections are a kind of explanatory connections. And we can also see what is
right about Street’s contention that moral facts do not “track” the moral facts. But rather than say
that there is no question-begging reason to think that there is a statistical correlation between
belief and fact, the problem is that non-natural moral facts don’t seem to be located in the right
place in the explanatory history of our moral evidence. (Natural moral facts might be so located
– this is why the Darwinian defeater is a problem for non-naturalism.
13
)

5. Debunking Third Factor Accounts
As we saw in Section 1, Third Factor Accounts create formidable problems for any of the
existing attempts to specify what makes evolution a defeater. So to conclude, I’ll show why
Third Factor Accounts are no help in answering the argument from the previous section.
Recall, Third Factor Accounts hold that, while our moral beliefs are not explained by the
moral facts, our moral beliefs are explained by something non-moral that is strongly correlated
with something moral, and this fact serves to dispel concerns related to the reliability of our
moral beliefs. In Enoch's example, facts about survival explain why we have the moral beliefs
that we have, which ensures that our moral beliefs reliably track facts about survival. But
survival is good; so our moral beliefs also track one aspect of the good, and thus are reliably
correlated with the moral facts. And as Berker (2014) and Vavova (2014a) argue, we are, in
general, entitled to make assumptions like this when answering the skeptic. If we are unable to

13
See Copp (2008).
115

make assumptions about the way the world is when responding to the skeptic, it will be
impossible to answer any skeptical challenge.
Third Factor Accounts can succeed at the task of providing an explanation of why our
moral beliefs might be reliably formed. But they cannot overturn an undercutting defeater. We
can show this by considering what would happen if we try to use a Third Factor Account to
counter another paradigmatic instance of undercutting defeat. Take the case of the Mischievous
Lab Assistant. Suppose the chemist were to reason in the following way: “I've just learned that
my lab assistant falsified my data. Should I stop believing that fire is phlogiston? Well, I don't
see why I should! After all, I haven't learned anything inconsistent with fire’s being phlogiston –
my lab assistant could have given me data that is actually reliable. So what do I think fire is? In
the context of this epistemic challenge to my belief, I am entitled to make some assumptions
about the nature of fire – I am dealing with an epistemic challenge that is supposed to have force
no matter what the fact of the matter is. So I'll assume that it is phlogiston. And what could be a
more natural assumption to make? After all, all my data supports the conclusion that fire is
phlogiston. So fire is phlogiston, my data says that it is, and my lab assistant invented the data. I
guess my lab assistant is really good at inventing reliable data. At any rate, my lab data is
reliable. Skeptical crisis averted!” The problems with this reasoning should be immediately
apparent. But this reasoning process is no different than the one Third Factor Theorists like to
walk us through when they reject Epistemological Debunking Arguments. The problems for
Third Factor Accounts are just as pressing.
Berker and Vavova are correct to say that we need to make some assumptions about the
way the world is in order to establish the reliability of some faculty. But we cannot maintain
116

those assumptions in the face of defeating evidence. We need to assume that the world is roughly
the way our senses portray it to be in order to answer the external world skeptic. But if we were
to learn that we are brains in vats, we would no longer be entitled to make that assumption.
Similarly, Enoch is right that we are entitled to assume that survival is good. But once we
encounter a defeater for that belief, we are no longer entitled to assume it in making an anti-
skeptical argument. And as I’ve argued here, there is a defeater for the belief that survival is good
– a Darwinian defeater.

6. Conclusion
When stating an Epistemological Debunking Argument, the devil is in the details. To
make a case against the possibility of moral knowledge, the debunker must rely on a correct
epistemic principle that shows clearly why the evolutionary history of our moral beliefs is so
worrying from an epistemic standpoint. I've argued here that evidence for a proposition will be
undercut whenever one learns that the evidence is explained by something other than the truth of
the proposition in question. This account of undercutting defeat has the potential to explain the
existence of defeat in a wide variety of cases in a natural and compelling way. And it has a strong
skeptical upshot for moral non-naturalism. Evolution is part of a larger, naturalistic story of why
we have the kinds of mental states that count as evidence for our moral beliefs. Because we have
good evidence in favor of this larger, naturalistic story, we have good reason to think that moral
facts play no role in explaining our moral evidence.
That's what makes evolution a defeater.
117

Chapter 5: Inferential Moral Knowledge and the Is/Ought Gap

1. Inferential Moral Knowledge
We turn now to the question of whether or not it is possible for anyone to have inferential
knowledge of non-natural moral properties. Inferential knowledge, recall, is knowledge whose
grounds include knowledge of other propositions, while basic knowledge is knowledge that is
not grounded in the knowledge of any other proposition. It follows that all knowledge is either
inferential or basic. In Chapter 3, I argued that basic moral knowledge is impossible. And in this
chapter, I will argue that inferential moral knowledge is impossible. The conclusions of these two
chapters together entail that, for the moral non-naturalist, all moral knowledge is impossible.
Since inferential moral knowledge is moral knowledge that requires other knowledge for
its grounds, we can now ask a further question: for a particular instance of inferential moral
knowledge, is it grounded in prior moral knowledge? Or is it grounded in prior non-moral
knowledge? While either is possible, in this chapter I will be solely concerned with moral
knowledge that is grounded in non-moral knowledge. Moral knowledge that is grounded in prior
moral knowledge solves no problems in moral epistemology, since we must then give an account
of how this prior moral knowledge is possible. (This is not to say that it is impossible for there to
be moral knowledge that is grounded in prior moral knowledge; if we have any moral
knowledge, then there is no reason why that moral knowledge may not ground further moral
knowledge.)
Putting things this way seems to presuppose a kind of foundationalism. Coherentists or
infinitists might object here that a lack of basic moral knowledge does not mean that inferential
moral knowledge must ultimately be grounded in basic non-moral knowledge. But this
foundationalist assumption is safe in this context. Infinitism, the view that the grounds for our
118

beliefs stretch back infinitely, is an absurd view. It is just not plausible that our grounds for
knowing something can stretch back, literally, infinitely. Given how inferential knowledge is
defined, this will mean that agents must have knowledge of an infinite number of distinct
propositions, each standing in a unique evidential support relation to another. That's implausible.
Coherentism is a much more reasonable view, but the plausibility of coherentism is a red herring
in the current context. Coherentism is, explicitly, a theory of justification, not a theory of
knowledge. And while a coherentist theory of justification is defensible, a coherentist theory of
knowledge is not. The central problem for coherentist theories of knowledge is the problem of
establishing that a method of relying on coherence relations is a reliable one. This is impossible,
because the coherence of a set of beliefs is no indication of the truth of those beliefs. There can
be coherent systems of beliefs that include only false beliefs, after all. Coherentists typically
respond to this charge of disconnection from the truth by arguing that theirs is a theory of
justification, not knowledge (see, e.g., Lehrer 1974). But this just concedes my point. I am rather
inclined to agree with the coherentist that coherence among one's beliefs can provide some
positive justificatory status to the beliefs that are thus related. But coherentism does not have the
structural features that would allow it to be extended to count as a theory of knowledge. Since
our business here is with knowledge, not justification, we will set coherentism aside.
I will evaluate the prospects for inferential knowledge by examining a number of
different kinds of inference, and seeing whether any of these varieties of inference will provide a
plausible route to inferring a moral proposition from a non-moral one. Section 2 will look at
deductive inference, Section 3 will look at inductive inference, Section 4 will look at abductive
inference, and Section 5 will examine the possibility of alternate inferential models. All four of
these approaches will be found wanting. Section 6 will conclude by reviewing the arguments of
119

Chapters 2-5 and showing how these arguments serve to support the first premise of the Moral
Closure Argument first presented in Chapter 1.

2. Deductive Inference
Deductive inference is a kind of inference where the truth of the premises guarantees the
truth of the conclusion. Good deductive arguments are deductively valid: it is impossible for the
premises to be true and the conclusion false. That much is common knowledge to all
philosophers, but this standard definition of validity has been interpreted in at least two different
ways by philosophers talking about valid arguments, and this difference of interpretation can
cause some confusion. The uncertainty surrounds the word 'impossible.' Some interpret
'impossible' to mean logically impossible,
1
i.e. there is no interpretation of the non-logical terms
involved in stating the premise and conclusion that makes the premises true and the conclusion
false. Others interpret it to mean metaphysically impossible, i.e. there is no possible world where
the premises are true and the conclusion is false.
2
On the former interpretation, the inference “X
is water, therefore X is H2O” will be an invalid inference. On the latter interpretation, that
inference is valid.
I will follow standard practice and give validity its logical, rather than metaphysical,
reading. The reason is that it is only the former kind of inference that is epistemologically useful.
Since the elimination and introduction rules for each of the logical connectives are, plausibly,
conceptually true of each of those connectives (e.g. anyone conceptually competent with
conjunction will see that 'A&B' can be derived from {A, B}), any individual who is conceptually
competent with logical terminology ought to be able to come to know that the conclusion of a

1
This is a common definition from logic textbooks, e.g. Boolos, Burgess, and Jeffrey (2007)
2
See Zimmerman (2010).
120

logically valid argument is true, if the premises are. Working with metaphysical validity, on the
other hand, introduces a host of epistemological complications. Someone's knowing that X is
water does not put them in a position to know that X is H2O. A related, and more serious, worry
arises when we consider instances of necessary truths. If the conclusion of an argument is
metaphysically necessary, then that argument will be metaphysically valid, no matter what the
premises are. But unlike logically necessary conclusions, not all metaphysically necessary
propositions are tautologies. The fact that water is H2O is metaphysically necessary, but that is
not the kind of thing that can be known on the basis of an argument without premises. More
generally, someone's knowing the premises of a metaphysically valid argument does not put
them in a position to know that the conclusion of that argument is true.
Since we are giving validity a logical reading, the prospects for giving a deductively valid
argument that begins with ‘is’ premises and ends with an ‘ought’ conclusion are going to look
rather slim. Formal logic is conservative: the conclusions of valid arguments cannot contain
more information than is contained in the premises. Hence, given an argument with only non-
moral information in the premises, there is no way to come to a conclusion that contains moral
information.
3

Despite this, some philosophers have tried to give counterexamples to the famous
Humean principle that an 'ought' cannot be derived from an 'is,' and some of these derivations
purport to give a deductively valid argument. One such argument comes from Prior (1960), who
argues thusly:
For any non-moral proposition R, and moral proposition M,
1. R entails (R v M)
2. (R v M) & ~R entails M.
3. Either (R v M) is a moral proposition or it is not.

3
See Pigden (1989) for an extensive discussion of this point.
121

4. If (R v M) is a moral proposition, then 1 is a logically valid argument with a non-
moral premise and a moral conclusion.
5. If (R v M) is a non-moral proposition, then 2 is a logically valid argument with two
non-moral premises and a moral conclusion.
6. Therefore, there is at least one logically valid argument with only non-moral premises
and a moral conclusion.
4

Prior's argument makes for a difficult puzzle for anyone who tries to maintain that there
can be no logical derivation of an 'ought' from an 'is.' The argument is valid, and each of its
premises seems to be beyond the slightest doubt. Most discussions of Prior's argument are
focused primarily on the logical structure of Prior's argument and the question of what it takes
for a proposition to have moral content. But we can set the question of “moral content” aside;
5

what matters for our purposes here is whether Prior's argument can have any epistemological
value.
Our goal in looking for a way to derive an 'ought' from an 'is' (at least in the context of
this dissertation) is not driven by an interest in the nature and limits of logical entailment; we are
trying to find out whether we can come to know a moral proposition on purely non-moral
grounds. And Prior's argument gives us no way of doing this. If we first come to know R, we are
indeed in a position to know that (R v M), but this does not put us in a position to gain
substantive moral knowledge. This would be like saying that learning that 2 + 2 = 4 puts us in a

4
This way of presenting Pryor's argument was taken from Maguire (forthcoming).
5
For those who are interested, here's my take on the logic of Prior: (R v M) is neither a moral proposition nor a
non-moral proposition. I believe that the best understanding of what it is for a true proposition to have “moral
content” and thus be a “moral proposition” is that a proposition has moral content if and only if the state of
affairs in the world that renders the proposition true is a moral state of affairs, i.e. it is (partly) constituted by the
instantiation of a moral property. Consequently, what makes a true proposition a “moral proposition” is not
intrinsic to the proposition itself, but is rather a function of the proposition and the state of the world that makes
it true. And there are two different states of affairs that can make (R v M) true. One is a moral state of affairs – M
is true – and the other is a non-moral state of affairs – R is true. So whether or not (R v M) is a moral proposition
will depend on whether or not (R v M) is true because of M or true because of R. If R is true, a trivial disjunction
introduction will not deliver a moral proposition. And if (R v M) & ~R is true, then what is making (R v M) true
must have been M to begin with. So in the first case, (R v M) is a non-moral proposition. In the second case, it is
a moral proposition. (Note that the two cases are mutually exclusive, since the first involves a commitment to R,
the second involves a commitment to ~R).
122

position to learn anything about astronomy. (Let R stand for '2 + 2 = 4' and M stand for 'the
number of stars in the sky is even.') It would be absurd to suggest that performing a trivial
disjunction after having learned a truth of basic arithmetic gives us something that might count
as astronomical knowledge. Similarly, if we know that (R v M) and we also know that ~R, we
can infer M. But if we know ~R and simultaneously know that (R v M), we cannot have any all-
things-considered good evidence in favor of R. Since we know that ~R, R cannot still be a live
possibility for us. It follows that our grounds for knowing that (R v M) must already suffice for
our knowing M. So our ability to perform a disjunctive syllogism here can't describe our method
for coming to know a moral proposition, since we will only be in a position to know the premises
of that syllogism if we already have moral knowledge.
The point generalizes to any attempt to provide a deductively valid argument from is to
ought. Because deductive logical inference is conservative, the conclusion cannot contain any
more information than is contained in the premises. What this means from an epistemological
standpoint is that one cannot begin with evidence that is only sufficient to know non-moral
premises, and then, through application of deductive inference, come to gain sufficient grounds
for moral knowledge.
There is one last interpretation of the notion of validity to consider before moving on.
This is an analytic notion – arguments are analytically valid just in case the conclusion can be
derived from the premises using only rules of logical inference and conceptual truths.
Accordingly, Bob is a bachelor; therefore, Bob is unmarried will be analytically valid, since the
bridge principle needed to derive the conclusion from the premise – all bachelors are unmarried
– is an analytic truth.
For an example of this kind of argument in the moral realm, consider the following
123

argument from Searle (1964):
(I) Jones uttered the words "I hereby promise to pay you, Smith, five dollars."
(2) Jones promised to pay Smith five dollars.
(3) Jones placed himself under (undertook) an obligation to pay Smith five dollars.
(4) Jones is under an obligation to pay Smith five dollars.
(5) Jones ought to pay Smith five dollars.
While this argument is not logically valid, it seems like each of the statements follows
from the statement preceding it as a matter of conceptual truth. So it seems like we can prove (5)
on the basis of (1).
Unfortunately for Searle, this argument equivocates on statement (2). (2) states that Jones
has performed a certain kind of speech act – an act of promising, but there are two different kinds
of speech act that (2) might pick out – an illocutionary act and a perlocutionary act. An
illocutionary act is an act of asserting some proposition or another, while a perlocutionary act is a
speech act that effects the world in some way. Utterances of sentences are actions that can be
both illocutionary and perlocutionary. Thus, for instance, the illocutionary act of saying “There's
a tiger!” consists in asserting something about the location of a certain large cat. The
perlocutionary act of saying the same thing consists (e.g.) in alerting someone about the
imminent danger. One can perform the former kind of speech act without performing the latter,
as when the listener does not hear the panicked shouts of the lookout. Similarly, an utterance of
the words “I promise to pay” can be both an illocutionary act and a perlocutionary act. The
illocutionary act is the act of asserting that you will pay. And the (relevant) perlocutionary act
performed by saying 'I promise to pay' is the act of placing oneself under an obligation to pay".
So if we understand (2) as claiming that Jones has performed an illocutionary act of promising,
124

then (2) follows from (1) as a matter of conceptual necessity. But (3) no longer follows from (2).
Searle himself recognizes this point in the closing passages of his paper, and argues on this basis
that “the alleged distinction between descriptive and evaluative utterances is useful only as a
distinction between two kinds of illocutionary force,” which he takes to be a shortcoming of a
narrow focus on the illocutionary force of speech acts. If we take (2) to be claiming that Jones
has performed a perlocutionary act of promising, then (3) does indeed follow from (2) as a
matter of conceptual necessity, since the perlocutionary force of promising something is to place
oneself under an obligation. But if we interpret (2) as a perlocutionary act, (2) no longer follows
from (1), since it is not, as a matter of conceptual necessity, impossible for one to utter the words
“I promise” without thereby placing oneself under an obligation.
The fact that Searle's argument fails should not be surprising. In order for there to be
analytic deductive inferences from 'is' to 'ought,' there will need to be analytic bridge principles
that specify when a certain naturalistically-specifiable state of affairs is sufficient for a moral
property obtaining. But as we saw in Chapter 2, there are no such analytic bridge principles.
I conclude that there can be no inferential moral knowledge when the inference-type in
question is deductive inference. Other kinds of inference might be more promising.

3. Inductive Inference
Inductive inference is inference from a series of past instances which conform to a pattern
to the conclusion that the pattern will continue in the next instance. Inductive knowledge of the
future is a paradigm instance of inductive knowledge – the sun has risen every day in my
experience (and many other days before I was around to experience sunrises), and on this basis I
can conclude that the sun will rise tomorrow. There is nothing special about extending patterns
125

from the past into the future, of course. I know firsthand that the sun has risen every day of my
life. This provides me with an inductive basis for concluding that the sun rose the day before I
was born.
Inductive inference also does not look like a promising candidate for providing us with
inferential moral knowledge. Inductive knowledge is obtained by noticing and then extending a
pattern. A history of past instances of purely non-moral knowledge cannot be extended by
induction to generate an instance of moral knowledge. Moral knowledge is different from non-
moral knowledge; moral propositions don't extend patterns of non-moral propositions. The only
way to obtain inductive knowledge of moral facts is by noticing a pattern of moral facts that can
then be extended. But this is inferential knowledge from moral premises, and that's not what
we're looking for here. Induction can't cross the is/ought gap.
To my knowledge, there has been only one attempt to show that induction can bridge the
is/ought gap, and that is from Mark Nelson (1995).
6
Nelson argues that we can have moral
knowledge by testimony, as in the following argument:
1. Everything Evan the Expert says is true.
2. Evan the Expert said that abortion is wrong.
3. Therefore, abortion is wrong
This argument is deductively valid, yet we should count it as an instance of inductive
knowledge because of the problem of how the first premise can be known. As we saw in the

6
Nelson has another paper, called “The possibility of inductive moral arguments,” in which he gives a different
“inductive” argument for moral knowledge. I use scare quotes here because Nelson has a different conception of
what it takes for an argument to be inductive – for Nelson, an inductive argument is any argument with a merely
probabilistic conclusion. Hence, he considers the following to be an inductive argument: It seems to me that M.
Therefore, probably, M. Nelson's case that this is a good argument rests on an endorsement of Phenomenal
Conservativism, together with some (dubious) auxiliary premises connecting the appropriateness of seeing that
one has a (potentially weak, potentially defeated) reason to believe that P and concluding “Probably, P.” I won't
go into the details, since I have already argued in Chapter 2 against trying to leverage Phenomenal
Conservativism into an account of moral knowledge. These two papers by Nelson give the only two attempts of
which I am aware to state an argument for inductive moral knowledge.
126

previous section, when evaluating the epistemic significance of arguments, we should examine
the agent's grounds for believing the premises. The dubious premise in this argument is Premise
1, but we need not include any moral propositions in our grounds for knowing that Premise 1 is
true. Premise 1 might be supported inductively by appeal to past instances of Evan saying true
things. With a sufficient track record of Evan saying true things, particularly if Evan is
demonstrating a keen intellect and speaking thoughtfully and deliberately, perhaps we can come
to be in a position to know that Evan is the kind of person who always gets things right, such that
if Evan says a thing, there is an overwhelming likelihood that it will be true. Thus, on the basis of
Evan's demonstrated correctness on a wide variety of non-moral matters, we can come to know
that what Evan says is true, even when Evan is asserting moral propositions. So goes the
inductive argument from testimony.
7

There are two problems with this argument. First, testimony is not the right kind of thing
to resolve fundamental problems in moral epistemology. If I am to know that abortion is wrong
on the basis of Evan's testimony, Evan himself must first know that abortion is wrong. The chain
of testimonial knowledge needs to end somewhere with a person who has figured things out for
herself. So we will still need an account of moral knowledge that is not testimonial.
8

Second, we cannot come to know that someone is correct in their opinions about moral
matters on the basis of their being correct about other matters because expertise is domain-
specific. If Evan makes a huge number of true claims about quantum mechanics, this can provide
us with a strong inductive basis for concluding that Evan's next claim about quantum mechanics

7
I am here assuming that knowledge by testimony is a kind of inductive knowledge. This assumption is made
only to streamline presentation. Readers who consider testimonial knowledge to be its own kind of inferential
knowledge are free to treat testimony in this way – nothing in my argument turns on testimony being a special
case of induction.
8
Several others have made note of this problem for moral knowledge by testimony. See, e.g. Sinnott-Armstrong
(2006).
127

will be correct. But this evidence cannot provide us with a strong inductive basis for concluding
that Evan's next claim about French literature will be correct. Evan's track record of accuracy
about quantum mechanics gives us strong reason to believe that Evan is well-versed in that
particular field. But since it is possible for Evan to be well-versed in quantum mechanics without
being well-versed in French literature (most physicists are like this), Evan's established expertise
does not generalize. This also holds when we are looking at fields that are closely related:
expertise in physics does not guarantee expertise in chemistry, and expertise in French literature
does not guarantee expertise in English or Italian literature. Even if we are confronted with an
individual who is a confirmed expert in all areas of non-moral knowledge, this gives us no basis
for concluding that this individual is an expert in moral matters as well.
One might object here that someone with a track record of saying true things that is as
good as Evan's must be very careful and intelligent, and thus Evan will be in a position to work
out the moral truth for himself.
9
The problem with this response is that it assumes that care and
intelligence are sufficient to put one in a position to work out the moral truth. But why make this
assumption? Care and intelligence are not sufficient to put one in a position to work out facts
about French literature, after all – one must first have read some French literature. And any
attempt to show that this assumption is warranted will just reintroduce our first objection to
moral knowledge by testimony: any account of how the careful and intelligent Evan could have
worked out for himself the nature of right and wrong will, ipso facto, be an account of non-
testimonial moral knowledge. Moral knowledge by testimony might be possible – but only if
some other variety of moral knowledge is possible.
I conclude that inductive inference is no more promising an avenue for inferential moral
knowledge than deductive inference.

9
See Sinnott-Armstrong (2006, Ch. 8).
128

4. Abductive Inference
The situation becomes more interesting when we look to abductive inference. Abductive
inference, or inference to the best explanation (IBE) is not conservative, nor does it rely on past
patterns of correctness. IBE allows us to make epistemic leaps, beginning with premises about
one domain and ending with a conclusion about some other domain. For instance, we can use
IBE to figure out how microphysical objects interact on the basis of macrophysical observations.
So we might hold out hope that IBE can allow us to epistemically break into the moral domain
(so to speak) from the outside.
The prospects for IBE providing moral knowledge look even more promising when we
consider models of mathematical or perceptual knowledge. In Chapter 2, we saw that some
models of perceptual and mathematical knowledge consider those kinds of knowledge to be
inferential rather than basic. Indirect realists about perception hold that we can only have basic
knowledge of our own experiences, and we can infer facts about the external world on the basis
of this knowledge of our experiences. The kind of inference that is relevant here is almost always
IBE.
10
Thus, V ogel (1990) argues that we can rule out skeptical hypotheses like the evil demon or
the envatted brain on the basis of the fact that they contain features that make them worse
hypotheses than realist hypotheses about the external world; the evil demon hypothesis is ad hoc,
and the envatted-brain hypothesis is not simple. Similarly, Bonjour (2003) lists a number of
phenomena that are pervasive in our visual experiences, like the fact that objects tend to get
smaller as we move away from them, and the fact that when one object is placed in front of
another, it occludes the object behind it. And according to Bonjour, the hypothesis that there are

10
Some hold that the kind of inference in question is a kind of sui generis inference. We'll look at sui generis
inference of moral facts in Section 5.
129

real objects that stand in spatial relations to one another is a uniquely elegant explanation of why
our visual experiences go through the patterns that they go through.
Similarly, the indispensability arguments that are popular in mathematical epistemology
are versions of IBE. Scientific theories are paradigm instances of abductive knowledge, and so
we gain abductive knowledge of everything that those scientific theories quantify over: both
scientific objects (like negative charges) and mathematical objects (like negative numbers). Since
scientific theories cannot be formulated without appealing to mathematical propositions (albeit
somewhat indirectly)
11
, mathematical propositions can be inferentially known on the basis of the
observations that these scientific theories predict. Consequently, the view that moral knowledge
can be known by IBE is the last, best hope for arguing that moral knowledge can be understood
as being relevantly similar to either perceptual or mathematical knowledge.
If we have abductive moral knowledge, this would entail that moral facts can explain
(and therefore be justified by) non-moral facts. So the possibility of abductive moral knowledge
turns on the possibility of providing moral explanations.

4.1 Moral Explanations
The view that moral facts have explanatory power has been a contentious one in
metaethics ever since Harman (1977) first cast doubt on the possibility of providing good moral
explanations. Harman famously argued that moral facts are not essential to the explanation of
any “observation;” Whenever any agent makes a moral judgment, this judgment can be
explained by invoking the attitudes of the agent without mentioning moral facts. Harman has

11
This isn't an uncontroversial matter. The nominalist project in the philosophy of mathematics, mentioned in
Chapter 2, attempts to formulate scientific theories without quantifying over mathematical objects. My
contention here is only that, if indispensability arguments in the philosophy of mathematics work, they do so by
appealing to IBE.
130

generally been read as offering up a version of an argument from Ockham's Razor, an ontological
principle that declares that we should leave our ontologies as sparse as possible, and therefore
should do away with any entities that serve no explanatory purpose. If moral facts explain
nothing, then they serve no purpose; they are ontological danglers, which can and should be
trimmed away through judicious application of Ockham's Razor. This is how Richard Joyce
reads Harman, when he writes “with moral judgments thus explained without recourse to moral
facts, and without anything else whose explanation requires us to posit moral facts... we have no
reason to believe that anything at all is morally wrong” (Joyce 2006, p. 186-187). Joyce later
adds that, unless moral facts can be reduced to natural facts with explanatory power, they can “be
excised from the picture with a swift slash from Ockham's Razor, since we have a complete
explanation of moral judgment with no need to posit any extra ontology in the form of moral
facts” (p. 188).
We can distinguish two views on moral explanations that one might defend in response to
Harman's argument. The first says that moral explanations are strictly better than non-moral
explanations. The second says that moral explanations are reasonable explanations that are just
as good as non-moral explanations. Call the former view Better Moral Explanations, and the
latter view Equal Moral Explanations. The second view is strictly weaker than the first view, and
thus easier to defend. And, fortunately for the moral realist, Equal Moral Explanations is all that
is needed to rebut Harman's argument from Ockham's Razor.
12
To see this, consider a case where
some set of data admits of two and only two mutually exclusive explanations, one of which
quantifies over the Fs, the other of which quantifies over the Gs. Assume for the sake of reductio
that if a type of object is not part of the best explanation of a set of data, then we can conclude
that no objects of that type exist. The Fs are not part of the best explanation of our data set, since

12
This is one of the main lessons to be drawn from Sturgeon (1988)'s response to Harman.
131

the F-involving explanation is no better than the rival G-involving explanation. So by Ockham's
Razor, we can conclude that there are no Fs. But by parity of reasoning, we can also conclude
that there are no Gs, since the G-involving explanation is no better than the F-involving one.
Thus, we are in a position to conclude that there are no Fs and that there are no Gs. But there
must be either Fs or Gs, since either the F-involving explanation of the data or the G-involving
explanation must be true. The conclusion to draw is that we cannot use Ockham's Razor to cut a
type of object out of our ontology merely on the grounds that explanations that entail the
existence of this object are no better than rival explanations. Thus, if Equal Moral Explanations
is true, Ockham's Razor will not serve to show that there are no moral properties. So if Equal
Moral Explanations is true, Harman's argument fails.
I take no stand here on whether Equal Moral Explanations is true. But as I will now
argue, even if Equal Moral Explanations is true, moral realists face a different, epistemological
problem, in place of the ontological problem from Ockham's Razor.
Let us again consider the case where two equally-good, mutually-exclusive explanations
exist for a set of data, one of which is F-involving, the other of which is G-involving. Assume
further that the explanatory power of these two theories is the only relevant epistemological
consideration in favor of either theory, or at least that all other evidence for and against either
theory is exactly balanced, and that it is common knowledge that these two explanations are
mutually exclusive. In such a situation, are we justified in believing that there are Fs? Clearly we
cannot be. If we were justified in believing that there are Fs, then, by parity of reasoning, we
ought to be justified in believing that there are Gs, since the evidence in favor of F-entailing
propositions is just as strong as the evidence in favor of G-entailing propositions. But we cannot
be justified in believing that both the Fs and the Gs exist. Ex hypothesi, the Fs exist if and only if
132

the Gs do not, and we know this. So we are not justified in believing that the Gs exist rather than
the Fs, or vice versa; we ought to be indifferent between these two theories. And since
justification is necessary for knowledge, we are not in a position to know of the true theory that it
is actually true.
It follows that, if Equal Moral Explanations is true, we are not justified in believing in
moral facts. If two explanations are equally good, we have no grounds for ruling out one or the
other by using Ockham's Razor. But so, too, are we not justified on the basis of IBE in believing
in one theory over the other. So while appealing to Equal Moral Explanations can make problems
for Harman's ontological argument that there are no moral properties, it cannot deliver the
stronger conclusion that we have moral knowledge by abductive inference.
13

This conclusion should not be surprising: if we are to know some moral fact by inference
to the best explanation, it really must be the best explanation that we are inferring to. But
emphasizing this point serves to clarify the dialectical burdens for the advocate of abductive
moral knowledge. In order to support the view that there is abductive moral knowledge, it cannot
merely be the case that moral facts figure into an explanation of some data or even a good
explanation of some data. They must figure into explanations that are better than any purely non-
moral explanation. So it is Better Moral Explanations that the anti-skeptic must defend if we are
to have abductive moral knowledge.
Looking at the situation this way also helps us counter a point that various moral realists
have made in response to Harman. Harman argues that moral facts can explain no natural facts,

13
The literature on moral explanations would be vastly improved by paying careful attention to the difference
between the ontological reading of Harman's argument and the epistemic one. Parties to this debate typically ask
questions like “Are we justified in believing in moral facts if those facts play no explanatory role?” But the
“justified in believing in” locution admits of both an ontological and epistemological reading. And as we've just
seen, the argumentative burdens can differ quite a bit depending on whether we are arguing for an ontological or
epistemological conclusion.
133

and for that reason we can do without them. But why restrict the range of the relevant
explananda to just natural facts? Moral facts can explain other moral facts, certainly; that kind of
explanation is not inconsistent with moral non-naturalism. If all that is required for a kind of fact
to be immune to Ockham's Razor is that it be capable of explaining something, then the ability of
moral facts to explain other moral facts should suffice for moral facts having full ontological
credentials.
14

While this might be an adequate line of response to Harman's argument, it fails to make
any progress in favor of abductive moral knowledge. When a fact explains some other fact that
we know to be true, the explananda counts as the premise(s) in the abductive argument for the
best explanation of that explananda. Hence, when moral facts explain other moral facts, this
explanation can only be construed as an abductive argument for a moral conclusion that also has
moral premises. So explanations of this kind aren't the right kinds of things to provide inferences
that cross the is/ought gap. Similarly, arguments that attempt to vindicate the ontological
credentials of moral facts by saying that they can enter into the explanatory economy of the
world because they are explained by natural facts, rather than the other way around,
15
are open to
the same objection. In abductive inferences, the only evidence is constituted by the explananda.
Consequently, the defender of abductive moral knowledge must hold that Better Moral
Explanations is true of natural explananda.
16

14
See Brink (1989, p. 182-183) and Enoch (2011, p. 51-52) for musings along these lines.
15
See Thomson's contribution to Harman and Thomson (1996), p. 93.
16
It is generally moral naturalists, in particular the “Cornell realists” like Sturgeon, who try to argue that moral
facts enter into the best explanations of natural facts, but non-naturalists have made this claim as well. Cuneo
(2006) endorses moral non-naturalism while arguing that moral facts have explanatory power. (A frustrating
wrinkle – Cuneo describes himself as defending moral naturalism. This is because Cuneo endorses the following
theses: (1) A property is a natural property if and only if it enters into causal relations. (2) Any kind of
explanatory relation is a kind of causal relation. (3) Moral properties enter into explanatory relations. Cuneo is
using his terms in a different way than I am; on my way of drawing the natural/non-natural distinction, Cuneo
counts as a non-naturalist. As I draw the distinction, natural properties are those with efficient causal powers. )
134

4.2 Badness and Baseness
Unfortunately for the non-naturalist, Better Moral Explanations is false. As Sinnott-
Armstrong (2006) argues, the fundamental problem for any attempt to give a moral explanation
of any particular datum that is superior to a non-moral explanation comes from the
supervenience of the moral on the non-moral. This moral supervenience is a kind of
determination or grounding relation. For an instance of bad pain, the badness of this event is not
(for the non-naturalist) constituted by the pain. But the event is bad because it is painful. More
generally, for any morally significant state, there is some state of affairs that is describable in
wholly non-moral terms that is sufficient for the moral state of affairs obtaining. This much is
uncontroversial.
Because of this, for any moral state of affairs, we can always (in principle) isolate that
specific non-moral state of affairs whose obtaining is sufficient for the moral property's
obtaining. It might be that we do not have words for these natural states of affairs, so that when
we consider all and only the morally-relevant factors, the only terminology that we have to use is
morally-laden. Yet the natural properties and the moral properties here are distinct, according to
the non-naturalist, and so we can always introduce a term (if needed) to pick out that wholly
natural state of affairs upon which the relevant moral properties supervene. Sinnott-Armstrong
suggests that we do just this, using the term “Base” to pick out the subvenient naturalistic basis
for any instance of the moral property of badness. (We can perform a similar operation to label
the subvenient naturalistic base of any moral property whatsoever – not just badness, but
goodness, rightness, wrongness, the counts-in-favor-of relation, etc. For purposes of simplicity,
we'll restrict our focus to Baseness and Badness).
Even moral nihilists can concede that Baseness exists. Baseness is entirely natural and
135

thus (for the non-naturalist) not constitutive of any moral property. So when it comes to the
question of whether moral properties provide superior explanations to non-moral ones, we
should understand the relevant non-moral explanations to be explanations in terms of Baseness,
but not Badness. It is then incumbent upon the moral non-naturalist to argue that explanations in
terms of Badness are superior to explanations in terms of Baseness. But there is no sense in
which this will be the case. Baseness has causal powers; Badness does not. Baseness can be a
constituent of or be constituted by other non-moral states of affairs; Badness cannot. Indeed, it
seems as though any attempt to provide a kind of moral explanation can, and indeed must be,
carried out in terms of the subvenient base property. To borrow a famous example from Sturgeon
(1988) – why did First Midshipman Woodworth fail to rescue the Donner Party? The easy
answer from Sturgeon is that Woodworth was no damn good. But his being no damn good
supervenes on the naturalistically-describable character dispositions that he has, and these
character dispositions have all the explanatory power that might be needed to explain any fact
about Woodworth. Indeed, it seems as though if Woodworth's Badness has explanatory power,
this is in virtue of the fact that his Badness supervenes upon his Baseness.
In light of these considerations, we may suspect that non-moral explanations really are
better than moral explanations, since explanations in terms of Baseness have as much
explanatory power as we might want, leaving Badness as an extraneous property. But even if we
think that adding moral properties like Badness to our explanations does not turn them into
worse explanations (perhaps proliferating objects is objectionable from the standpoint of
ontological parsimony, but proliferating properties is not?), adding Badness to an explanatory
story that is already committed to the existence of Baseness does nothing to enhance the
simplicity, explanatory power, or cohesiveness of the explanation. Therefore, Better Moral
136

Explanations is false. So for the non-naturalist, abductive moral knowledge is impossible.
Majors (2007) objects to this argument on the grounds that moral properties are multiply
realizable. Focusing on the case of injustice, rather than Badness, Majors says of the subvenient
base of injustice, “the closest possible world in which this set of naturalistic properties is not
instantiated is a world in which a different set was present, one which is alike sufficient for the
instantiation of injustice. Since injustice passes the causal-explanatory test, and no non-moral
property does, it follows that moral properties do explanatory work that cannot be done by their
subvenient bases” (p. 10). Majors' point is that, since the moral property of injustice can be
instantiated in any number of different natural states of affairs, explanations in terms of natural
properties have less explanatory power than moral explanations. This is because the “best
explanation” is one that describes the explanans at the level of generality most suitable to our
explanatory purposes. If Badness is realizable in any number of distinct natural states of affairs,
and I will judge that something is bad whenever it is bad, then the best explanation of why I
judge that thing to be bad will be explanation in terms of the moral property, which is present in
every case of my judgment, rather than the natural property, which is not.
17

Majors's argument rests on the assumption that Baseness is not multiply realizable. But
this assumption is unwarranted. Baseness, as I stipulatively define it here, is a multiply realizable
natural property, which will be instantiated in any instance where Badness is purportedly
instantiated. The fact that Baseness is multiply realizable means that many acts with different
(fine-grained) natural properties can instantiate Baseness, and hence explanations in terms of
Baseness can provide explanations at the right level of generality to count as the best
explanations.
Importantly, Baseness is not anything “over and above” the natural properties. It is

17
See also Oddie (2005, Ch. 6-7) for an extensive development of this idea.
137

multiply realizable in any number of different natural states of affairs, but in each instance where
Baseness is instantiated, it is wholly realized in a natural state of affairs. Baseness is causally
efficacious as well. As Majors points out, many slightly different configurations of the natural
properties can have the same causal powers to make us act in certain ways or form certain moral
judgments. It is this causal power that unites all of these (otherwise distinct) natural states of
affairs. And it is this causal power that grounds the fact that each of these states of affairs is a
realizer of Baseness. All and only Base states of affairs have the explanatory power that realists
like Majors want to attribute to Badness; having these explanatory powers is constitutive of
Baseness.
(Instances of) Baseness are wholly constituted by natural properties, and Baseness is
causally efficacious. For the non-naturalist, moral properties are neither of these things. It
follows that, for the non-naturalist, Baseness cannot be identical to Badness. This means that, if
explanations in terms of Baseness are the best explanations that there are, then Better Moral
Explanations is false. I conclude that moral knowledge of non-natural properties via abductive
inference is impossible.

5. Alternative Inferential Models
In the last three sections, we have looked at three varieties of inference – deductive,
inductive, and abductive, and concluded that none of these kinds of inference can provide us
with moral knowledge. Perhaps the thing to do at this point is think outside the box. The three
standard forms of inference cannot provide us with inferential moral knowledge, but perhaps
some other kind of inference can. Perhaps some sui generis moral or practical inference can
provide us with moral knowledge on the basis of non-moral premises.
138

Thomson (1990) provides us with an example of an argument that looks like this.
Thomson presents a case where a certain individual has been hooked up to his doorbell in such a
way that pressing the doorbell will cause this individual great pain. Thomson then presents the
following argument:
(1) If C rings D's doorbell he will thereby cause D pain.
Therefore,
(2) Other things being equal, C ought not ring D's doorbell.
As before, it is hard to see how (1) could be true and (2) false.
Let's get clear on how not to interpret this argument. This is not a deductive argument that
has made a rather poor attempt to be formally valid, which the principle of charity will require us
to modify by making some suppressed premise explicit (i.e. an enthymeme). This argument is
not implicitly relying on a premise stating “If (1), then (2)” (or anything similar). For if such a
premise were to be added, then the argument would be deductively valid, and all of the problems
that arose for deductive inference to a moral conclusion that we looked at in Section 2 will arise
in full force. Specifically, we could rightly question how we could know the suppressed premise,
since that premise is itself a moral principle.
18
A better way to think of this argument might be to
say that the fact that (2) follows from (1) is essential to the structure of the argument, but not as a
premise. 'From (1), conclude (2)' should be taken as an inference rule, serving the same role in
the argument that modus ponens would if the suggested conditional premise were to be added.
The possibility of inferences like this is obviously not restricted to Thomson's doorbell
case. We could hold, quite generally, that there is an inference rule permitting inferences to the
existence of moral properties on the basis of those natural properties on which moral properties

18
Zimmerman (2010) and Sinnott-Armstrong (2006) both read Thomson as having given an enthymeme, and both
criticize Thomson's argument on the basis of the fact that the suppressed premise is in need of justification.
139

supervene.
The idea that there are inference rules that permit one to infer that a property obtains on
the basis of its subvenient base obtaining is a tempting one. And it is certainly true that if one
property does genuinely supervene on another property, the supervenient property will
necessarily be instantiated if the subvenient one is. But as always, it's a mistake to read too much
into the metaphysics of the situation when doing epistemology. Something's being H2O
supervenes on its being water, but that does not mean that we can know that something is H2O
on the basis of its being water without first being in a position to know that all water is H2O.
And metaphysically necessary truths trivially supervene on everything. The problems that we
noted in Section 2 for a metaphysical reading of the notion of validity are back in full force.
The thing to say here might be that the ability to infer that moral properties obtain on the
basis of their subvenient base obtaining does not follow from a general rule regarding what we
may infer when one property supervenes on another. Instead, this might be a kind of sui generis
moral inference, only applicable within the moral realm. While such a move would remove
overgeneration concerns to the effect that it will be too easy to come to know any necessary
truth, it does so only at the cost of making the view look unacceptably ad hoc. We are owed an
account of what it is about these moral inferences in particular that makes reasoning in
accordance with them epistemically virtuous.
An advocate of sui generis moral inference might abandon the claim that supervenience
is what matters, and instead try to focus on the psychological reality of the fact that we do
perform inferences like Thomson's inference. Perhaps the fact that we do perform inferences like
this gives us some reason to believe that this inference is reliable? However, this is not generally
the case. There are a large number of unreliable reasoning processes that people unfortunately
140

have a tendency to engage in. People will frequently affirm the consequent (I speak here mainly
from my experiences teaching basic logic), and will often believe that the probability of a
conjunction is higher than the probability of either of the individual conjuncts. The reason that
people make these fundamental mistakes in reasoning seems to be that they are applying some
mental heuristic, the overall adoption of which is beneficial, but which won't always get them to
having true beliefs (Kahneman 2011). And this same set of considerations might apply in the
moral case as well. It is advantageous to engage in moral patterns of reasoning, and this will be
true whether or not there are moral facts (Joyce 2006). So the psychological reality of a tendency
to reason in certain ways will not be evidence that this kind of reasoning is reliable.
While it's likely not the case that we need to know that an inference rule is reliable in
order to successfully use it, it seems to me that something weaker is true: we need to be in a
position to know that an inference rule is reliable in order to gain knowledge by applying that
inference rule, where an agent is in a position to know a proposition just in case the agent has
enough evidence to know that proposition, were she to base her belief on that evidence (i.e. she
is propositionally justified in believing it). And it seems to me that (almost) everyone is in a
position to know that deduction, induction, or abduction are reliable rules of inference. The truth-
preserving nature of deductive inference rules is a conceptual truth, since deductive inferences
follow the introduction and elimination rules for each of the logical connectives, and it is quite
plausible that those rules express conceptual truths about those connectives. The truth-preserving
nature of inductive inference and abductive inference do not seem to be conceptual truths. Yet
they are capable of independent empirical confirmation. Anyone older than the age of, say, two,
has enough experience with the world to recognize that events in the world occur in accordance
with regular patterns, and is thereby in a position to know that these patterns are something that
141

can be relied upon when forming beliefs about the future (e.g.). Similarly, anyone with a
sufficient amount of experience of the world is in a position to know that events have
explanations, and that the explanations for those events is usually the simplest, unified
explanation that can explain all of the data that needs to be explained. So the patterns that
induction and abduction rely on can be known about. Inductive inference and abductive
inference also yield results for which we can receive independent confirmation. When I conclude
on the basis of inductive inference that the sun will rise tomorrow, I only have to wait a day and I
can know first-hand that my inductive practice got things right. Similarly, when I conclude on
the basis of IBE that a machine works a certain way (e.g.), I need only open up the machine to
confirm that I was right.
In short, it is possible to have basic knowledge of propositions that I first come to believe
on the basis of either inductive or abductive inference, and this constitutes independent
confirmation of the reliability of these inference forms. But the same is not true of moral
propositions; as I argued in Chapter 3, basic knowledge of moral propositions is impossible. It
follows that sui generis moral inference cannot be independently confirmed, and is therefore
suspect.
Of course, not every fact that we can know of through inductive or abductive inference is
one that can be independently confirmed. But inductive and abductive reasoning processes are,
quite generally, reasoning processes that can and do enjoy a lot of independent support. And once
we have a certain amount of experience of the world, we are in a position to see that they do
enjoy a lot of independent support. This is why we are justified in relying on inductive and
abductive reasoning processes, but not sui generis moral inferences.
142

Chapter 6: The Relevance of Moral Skepticism
Throughout this dissertation, I've been examining ways to respond to the Closure
Argument for moral skepticism:
Moral Closure Argument
1. I do not know that I am not in the Natural World.
2. If I know that killing is wrong, then I know that I am not in the Natural World.
3. Therefore, I do not know that killing is wrong.
The argument generalizes to eliminate knowledge of any moral proposition. In previous
chapters, we've examined various strategies for rejecting Premise 1 of this argument, and found
them all to be lacking. This chapter will examine another kind of anti-skeptical strategy: a
“Relevant Alternatives” strategy. Some versions of the Relevant Alternatives strategy are ways of
rejecting Premise 1 of the Moral Closure Argument. Others are ways of rejecting Premise 2
(more on this in a minute). But what these strategies have in common is a commitment to the
thought that our inability to rule out a given skeptical hypotheses isn't troubling. This is because,
for the Relevant Alternatives theorist, not all skeptical hypotheses are “relevant;” they are not the
kinds of possibilities that one need rule out in order to have knowledge.
The bulk of the chapter will be dedicated to finding a neutral scheme for determining
what makes an alternative relevant or irrelevant. I will not try to motivate the relevant
alternatives strategy over any other anti-skeptical strategy; I am neutral on whether the relevant
alternatives strategy is a promising way to answer the skeptic. But, on the assumption that this is
a viable anti-skeptical strategy, we will see what kinds of things a relevant alternatives theorist
must say. And once we have a viable account of relevant alternatives in hand, we will see that a
Relevant Alternatives view is no help to the moral anti-skeptic. The Natural World hypothesis is
143

a relevant alternative.
Section 1 will examine the Relevant Alternatives account in detail and show how this
might work to disarm skeptical worries. Section 2 will examine the question of what it takes for
an alternative to be relevant, looking at a number of influential accounts, such as Contextualism
and a nearby-possible-worlds account, as well as some novel accounts, and find them all lacking.
But the failures examined in Section 2 will suggest a better account of relevant alternatives,
which I will present and defend in Section 3. Section 4 will then argue that, on this account, the
Natural World is a relevant alternative. This will conclude my defense of the Closure Argument
for moral skepticism.

1. Closure, Ruling Out, and Relevant Alternatives
What distinguishes a relevant alternatives anti-skeptical strategy from other kinds of anti-
skeptical strategies is an abandonment of the idea that we can prove that the skeptical hypothesis
does not obtain. The relevant alternatives account begins with what Duncan Pritchard has called
the “core relevant alternatives intuition:”
(RAI) In order to know a proposition, p, what is required is that one is able to
rule out all those not-p alternatives that are (in some sense to be specified)
relevant.
RAI is opposed to what we might call an “all alternatives” account of knowledge, which
says that it is a necessary and sufficient condition on knowledge that one be able to rule out any
alternative whatsoever. There are several different accounts of what it takes to rule an alternative
out, and for the purposes of this chapter I want to remain neutral between them. But we can say
at least the following: S can rule out an alternative A only if S has at least some internally
144

accessible evidence that supports ~A rather than A (we might call this “discriminating evidence;”
supporters of relevant alternatives frequently talk about the ability to discriminate).
1
Given this,
we might well worry that we are unable to rule out skeptical hypotheses, since skeptical
hypotheses are designed so as to conform to all of our evidence. So if having discriminating
evidence is a necessary condition on ruling out an alternative, and being able to rule out a
relevant alternative is a necessary condition on knowledge, and we don't in fact have any
evidence that discriminates between skeptical and non-skeptical hypotheses, then either skeptical
hypotheses are irrelevant or else we have no knowledge. This is why relevant alternatives
theorists tend to believe that skeptical hypotheses are irrelevant.
Discussions of relevant alternatives are intimately tied up with discussions of the Closure
Principle. As we saw in Chapter 1, the Moral Closure Argument is best understood as being a
compressed version of the Expanded Moral Closure Argument, of which the second premise
was:

Closure: If S knows that P, and P entails Q, and S believes that Q on the basis
of S's belief in P, while retaining knowledge of P throughout his reasoning,
then S knows that Q.
2

If Closure is false, the Moral Closure Argument is unsound. But the intuitive case in
favor of Closure is very strong. It is possible, after all, to expand one's knowledge through

1
By speaking here of “internally accessible evidence,” I am excluding certain externalist conceptions of evidence,
such as the theory of evidence advanced by Williamson (2000). This is because of the importance placed by
relevant alternatives theorists on the ability to rule something out or the capacity to discriminate. These notions
are agent-centric, and thus these capacities should supervene on resources that are available to the agent. To be
clear, I am not arguing here that a Williamsonian conception of evidence is false (although I admit I am not
sympathetic). Rather, the point is that advocates of relevant alternatives are generally hostile to a Williamsonian
conception of evidence. After all, if a Williamsonian conception of evidence is correct, agents do have the
resources to rule out skeptical alternatives. But relevant alternatives theorists do not try to resist skeptical
arguments in this way.
2
There are other ways to formulate Closure, and exactly what makes for the best version of the Closure Principle
is a matter of some controversy. What is offered here constitutes a very common version of Closure.
145

competent inference. And, in fact, it seems like we do this all of the time. If Closure is false,
then, it must be because it does not hold in full generality, not because there is something
mistaken about the idea that we can gain knowledge through the use of logic.
Many relevant alternatives theorists adopt exactly this stance. For instance, Dretske
(1970) held that different propositions can have different sets of relevant alternatives, and what
counts as a relevant alternative for two different propositions, P and Q, determines whether
knowledge of P yields knowledge of Q by Closure. If P and Q have the same set of relevant
alternatives, knowledge of P involves the ability to rule out all the relevant alternatives to P. And
if all of the relevant alternatives to Q are also relevant alternatives to P, then being able to rule
out all the alternatives to P entails being able to rule out all the alternatives to Q. So the Closure
Principle applies. But, if P has a smaller set of relevant alternatives than Q, then being able to
rule out all of the relevant alternatives to P does not entail being able to rule out all of the
alternatives to Q.
3
Thus, Closure will fail to hold when Q has a larger set of relevant alternatives
than P, and S is in a position to rule out only the propositions that are alternatives to P.
This model can be used to combat the skeptic. Say P is some common-sense proposition
(e.g. that I have hands) and Q is the denial of some skeptical hypothesis (e.g. that I am not a
handless BIV). If the class of relevant alternatives to P is smaller than the class of relevant
alternatives for Q, then I can know that I have hands even if I don't know that I am not a handless
BIV . Therefore, the second premise of the closure argument will be false, and the skeptical
conclusion can be avoided. Relevant Alternatives theorists like DeRose (1995) and Lewis (1996)
hold views of this kind.
But while Dretske, DeRose, and Lewis all want to reject Closure, other Relevant

3
I here assume that relevant alternatives are proposition-relative, not agent-relative or context-relative. This is
only for ease of exposition; we'll examine contextualist accounts of relevant alternatives in Section 2.1.
146

Alternatives theorists want to say that Closure is true in full generality. For instance, Stine (1975)
argues that all propositions have the same relevant alternatives, and so Closure is always true.
But the skeptic can still be defeated because skeptical hypotheses are always irrelevant. Thus,
one can know that skeptical hypotheses are false simply by believing that they are. This is
because all alternatives to the belief that the skeptical hypothesis is false will be versions of the
claim that the skeptical hypothesis is true, and none of these alternatives need to be ruled out
because they are irrelevant. This is a Moorean position in that it denies Premise 1 of the Closure
Argument, but is still distinctively a version of a relevant alternatives theory. Stine wants to say
that the skeptic is defeated not because we can safely rule out the skeptical hypotheses, but
because the skeptical hypotheses being offered need not be ruled out. So the Relevant
Alternatives theorist may or may not deny Closure. Closure-denial is not essential to the essence
of a Relevant Alternatives anti-skeptical strategy. What is essential is the claim that skeptical
hypotheses are not relevant. In order to make good on this general strategy, then, we need an
account of what makes an alternative to a proposition relevant. Investigating this question will be
my task in the next two sections.

2. What Relevant Alternatives Are Not
While there are a fairly large number of philosophers who support the Relevant
Alternatives view in one form or another, there is precious little agreement on what makes an
alternative relevant in the first place. In this section, I will go through a number of possible
candidates for what makes an alternative relevant, and show why they won't work. This will
direct us toward an acceptable account of what makes an alternative relevant: For an alternative
147

to be relevant, it can't have any of the problems that these other views have. By compiling a list
of problems, we will thereby be in a position to compile a set of criteria that any acceptable view
must meet. In the next section I will propose an account of what makes an alternative relevant
that can meet these criteria. The fact that it can do so will then count as evidence in favor of that
view.
2.1 Contextualism
There is a sharp divide in the literature between two different types of relevant
alternatives accounts: contextualists and invariantists. Invariantists say that the relevant
alternatives are a function of the proposition in question and the epistemic state of the agent.
Contextualists, on the other hand, focus not (exclusively) on the proposition and the agent, but
rather the context in which knowledge of that proposition is being ascribed to the agent. To a
contextualist, relevant alternatives are a linguistic phenomenon. The relevant alternatives to P –
and, hence, whether or not S's belief that P amounts to knowledge – will depend on the context
for an utterance of the sentence 'S knows that P.' That is, it's not just that the truth of 'S knows
that P' can change depending on changes in the status of S or P, but the truth-conditions of that
sentence can change as well, depending on its context of evaluation. Thus, sentences of the form
'S knows that P' will have different sets of relevant alternatives in different contexts. In “low
standards” situations a relatively small class of alternatives will count as relevant, while in “high
standards” situations a relatively large class of alternatives (including, perhaps, skeptical
alternatives) will count as relevant.
4

The contextualist treats knowledge attributions similarly to claims about tallness. In
contexts where I am talking about third-graders, being five feet high will count as being tall. In

4
Not all contextualist views are explicitly formulated as relevant alternatives views. Some are – DeRose, in
particular, models contexts as changing the set of alternatives that counts as relevant. But others are not.
148

contexts where I am talking about professional basketball players, being five feet high will count
as being not tall. So I can say of the same person “He is tall” (when talking about third-graders)
and “He is not tall” (when talking about basketball players), and say something true both times.
Similarly, according to the contextualist about knowledge ascriptions, I can say “Bob knows that
he has hands” when I am speaking in a quotidian context, and “Bob does not know that he has
hands” when I am in the epistemology seminar room, and say something true both times. The
difference in context shifts the standards for the knowledge attribution, and thereby shifts the set
of alternatives that count as relevant. The person on the street may well know that she has hands.
But in the epistemology seminar room, external world skepticism may in fact be true.
Contextualism has enjoyed a good deal of popularity in recent decades, but it faces some
very serious problems. I will present what I take the strongest challenges to contextualism to be,
not because these challenges are novel – most of them are taken directly from Hawthorne (2004)
– but in order to show why I believe contextualism should be set aside in favor of an invariantist
account of relevant alternatives.
The first problem with contextualism is that it's doubtful how well contexutalism fares as
a response to skepticism. In the philosophy seminar room, the skeptic's claims are all true,
according to the contextualist! While this result might do justice to our conviction that skeptical
arguments are hard to refute, the contextualist tells us that skeptical arguments are actually
impossible to refute
5
– the best we can do is get ourselves in situations where we can ignore the
skeptic. This is hardly the strong anti-skeptical conclusion we might have hoped for.

5
It is consistent with contextualism to hold that one can rule out skeptical alternatives in both high-standards and
low-standards cases. But it is very common for contextualists to adopt the position mentioned here, that most
knowledge claims are true in everyday contexts but false in high-standards contexts, because this is supposed to
do justice to both our everyday conviction that we know many things about the world, while also explaining how
strong a pull skeptical arguments can have on us in the seminar room. See Lewis (1996) and DeRose (1995) for a
development of this line of reasoning.
149

The larger problem for contextualism is that the contextualist view predicts certain
linguistic data that we don't actually find. For one thing, the contextualist predicts that when two
people make seemingly contradictory knowledge ascriptions in two different contexts, there will
be little or no pressure to say that they are contradicting each other. Compare: when discussing
heights, an assertion of “Timmy is tall” in a context where we are discussing third-graders will
sound true (Timmy is five feet tall), while in a context where we are discussing professional
basketball players, “Timmy is tall” will sound false. These statements require no qualification in
order to sound true or false, respectively, and if they are juxtaposed, anyone will easily be able to
say that there is some sort of context change that is making the difference (“Oh, well, that time I
was talking about basketball players!”). But the same is not true of knowledge ascriptions. An
assertion of “Timmy knows that he has hands” in quotidian contexts will sound true, and an
assertion of “Timmy knows that he has hands” in the philosophy seminar room still sounds true.
G E Moore has been read many times in philosophy seminar rooms, and while we might have
some discomfort with Moore's argument, that discomfort comes from the fact that it seems
question-begging or unwarranted in some way, not from a conviction that Moore didn't actually
know his key premise.
A contextualist might resist this point, saying that in the philosophy seminar room we are
much less likely to ascribe knowledge to Moore that he has hands. But this tendency is
accounted for not by a shift in context, but rather by the fact that in the seminar we confront
strong skeptical arguments, and confronting strong arguments that ~P is the kind of thing that
makes us doubt that P. We need not be contextualists about any term found in 'P' in order to
account for this. Consider: we are all typically satisfied that we and everyone that we know are
not moral monsters. Yet when we read Peter Singer's “Famine Affluence and Morality,” and feel
150

the pull of his arguments to the effect that everyone who does not give all of their wealth to
charities is a moral monster, we will begin to doubt our prior conviction that we are not actually
moral monsters. This observation about our response to Singer should not make us contextualists
about the term 'moral monster.'
While contextualism offers an elegant account of why we take skeptical hypotheses
seriously in some contexts but not others, it's not at all clear that the truth-conditions for our
knowledge ascriptions really do shift in the way the contextualist predicts when we move from
quotidian contexts to the philosophy seminar room. This is particularly evident when we
consider the case of moral knowledge attributions. In my experience, moral realists are realists in
the seminar room as well as outside it (my time in seminar rooms would have been much easier
were this not the case!) Being in a context where moral skeptical arguments like the Closure
Argument are being put forward does not cause one, ipso facto, to admit that one no longer know
that killing is wrong. But this is precisely the shift that would need to take place if the
contextualist is to provide a way out of the Closure Argument for Moral Skepticism.
Additionally, knowledge claims do not function in the way the contextualist view would
predict when it comes to retraction or disquotation.
6
As to disquotation, it seems like we do
accept a kind of disquotation schema for 'knows': Roughly, “When a speaker asserts a sentence
of the form “S knows that P,” that speaker believes that S knows that P.” On a contextualist
account, this disquotation schema will only seem true if the standard for knowledge that we are
using when we are talking about the speaker's beliefs is the same as the standard for knowledge
that the speaker was using when he made the assertion in question. Of course, there is no
guarantee that our standards will be the same as the speaker's standards. So according to the
contextualist, this disquotation schema will often come out false.

6
These objections come from Hawthorne (2004). As such, I only sketch these objections here.
151

While there are, perhaps, contexts in which the disquotation schema does sound fishy,
7
it
should again be noted that the high standards present in discussions of skepticism do not make
the disquotation schema sound fishy. If a person on the street believes “I know that I have hands”
and “I know that killing is wrong,” I can accurately report her beliefs by saying “she believes
that she knows that she has hands and she knows that killing is wrong,” even if I am in the
seminar room discussing both external world and moral skepticism.
The contextualist also gets data about retraction wrong. When we make an assertion in
one context, and then our assertion is later brought up and challenged in another context, if our
assertion was regarding some contextually variable standard, we will often clarify our earlier
assertion – “I only claimed that Timmy was tall for a third-grader.” But we don't say things like
this regarding our knowledge claims. In the seminar room, if someone points out that I earlier
claimed to have hands, I will either try to defend my earlier assertion or concede that I was
wrong. I won't try to say something like “I only claimed that I knew for every-day contexts.”
According to the contextualist, there is a change in the truth conditions of knowledge
attributions between quotidian contexts and “high standards” contexts where skeptical arguments
are taken seriously. But while we might feel more pull from skeptical arguments when we are
taking them seriously, we can account for this increased pull by simply noting that we are in a
context where we are taking this argument seriously. We do not encounter any of the linguistic
data that we would expect if the standards of knowledge actually did change in these high
standards linguistic contexts. Yet it is precisely this kind of change that would be needed in order
to vindicate the contextualist response to a skeptical Closure Argument.
None of these objections are intended to be either novel or decisive. But the problems

7
Perhaps DeRose's famous “Bank” case is one of these. However, I do not believe that our intuitions about the
Bank case provide any support for contextualism – see Lutz (2014).
152

with contextualism are extensive enough that we should see whether invariantist accounts can
provide us with everything we need from an account of relevant alternatives.
2.2 Truth in Nearby Possible Worlds
The most popular Closure-denying version of the Relevant Alternatives view is one
according to which an alternative is relevant just in case it obtains in nearby possible worlds.
According to Fred Dretske's original 1970 formulation of the Relevant Alternatives view, “a
relevant alternative is an alternative that might have been realized in the existing circumstances if
the actual state of affairs had not materialized” (Dretske 1970, p. 1021). Duncan Pritchard (2010)
defends a version of this view specifically formulated in terms of “nearby possible worlds.”
8
And
because this is the original formulation of the relevant alternatives view, and it still has
contemporary advocates, this kind of view deserves our close attention. According to Dretske
and Pritchard, if some proposition P is true in this world, but in nearby possible worlds, P is false
and Q is true instead, then for an agent to know P, that agent must be able to rule out the
possibility that Q.
This account of what it takes to have knowledge is formulated in order to give an account
of perceptual knowledge. If I see a black bird and believe that it is a crow, while in nearby
possible worlds the black bird that I am looking at is a raven, I need to be able to rule out the
possibility that the bird I am looking at is a raven in order to have knowledge that it is a crow.
My possession of knowledge about whether or not a certain bird is a crow will therefore turn on
my ability to discriminate between ravens and crows. And this, it seems, is as it should be.
Conversely, if an alternative would not obtain in any nearby possible world, I need not rule it out.
This is the underlying thought behind Dretske's famous example of the cleverly-painted mule.
None of my visual evidence discriminates between zebras and cleverly-painted mules (at least

8
So do several others. See, e.g. Heller (1999)
153

not for me), yet I can nonetheless know that the animal I am looking at is a zebra because I am in
a reputable zoo, and there are no nearby possible worlds in which cleverly-painted mules occupy
the zebra cages in reputable zoos.
This account of relevant alternatives in terms of nearby possible worlds raises an
interesting and important question – what does it take for a world to be nearby? The usual
assumption among those who favor a nearby possible worlds account is some sort of
metaphysical or nomological proximity, as articulated by David Lewis. According to this
account, a world is nearby if and only if the nearby world would have been the actual world, had
a small divergence miracle occurred relatively recently. For instance: Had I looked left instead of
looked right, I would have seen a raven rather than a crow. Thus, there is a nearby possible world
in which I'm looking at a raven – that's why the proposition that the bird I am looking at is a
raven is relevant. While this account of knowledge in terms of metaphysically nearby possible
worlds seems to get the right results in cases of perceptual knowledge, it is a very poor general
account of knowledge, since it gets very bad results when we are dealing with knowledge of
necessary truths. For necessary truths, of course, there are no nearby possible worlds where the
necessary truth in question is false and some other alternative is true. The truth in question will
be true in all possible worlds; accordingly, there will be no worlds in which any alternative to a
necessary truth is true. And so the relevant contrast class for any necessary truth will be empty,
since there are no alternatives that are true in nearby possible worlds. As such, the requirement
that an agent must be able to rule out all possible alternatives in order to have knowledge will be
trivially satisfied.
This is a bad result. For one thing, it seems to make knowledge of necessary truths far too
easy to come by. Take the fact that water is H2O, and consider Lavoisier, who is just beginning
154

to conduct experiments to determine the chemical composition of water. Before conducting any
experiments, Lavoisier has no evidence that can discriminate between the possibility that water is
H2O or the possibility that it is an element, or perhaps some other chemical compound. In this
case, it seems like, intuitively, there are relevant alternatives to the truth of that identity claim –
in particular, that water might be an element or another compound. That Lavoisier cannot yet
rule out any of these possibilities counts strongly against his knowing that water is H2O, even if
he happens to believe it.
A more general problem that is raised by this kind of view is that, on this version of a
relevant alternatives view (and perhaps several others), some propositions will have no relevant
alternatives so long as the proposition is true. But that's not what we want out of our account of
relevant alternatives. The relevant alternatives intuition isn't just a restatement of the factivity of
knowledge, it's a way of identifying the alternatives to a given proposition that our evidence
needs to be able to deal with.
The obvious response to make to these kinds of objections is to say that I'm focusing on
the wrong kind of modality. White it is metaphysically or nomologically impossible that water be
something other than H2O, there might be relevant counter-possibilities that count as nearby
possibilities on some other conception of possibility. That water is an element might be an
epistemically nearby possible world for some agents, and perhaps we should consider
alternatives to be relevant just in case they are true in worlds that are epistemically nearby.
The main problem with this suggestion is that it is incomplete. Speaking in terms of
metaphysical or nomological closeness is helpful because we have, from Lewis, an account of
what it is for a world to be nearby in this sense. But talk of “nearby possible worlds” is nothing
more than an evocative way of introducing a formal framework. Saying that the relevant
155

alternatives are just the alternatives that are true in nearby possible worlds is just to say that there
is a (partial) ordering of worlds such that the relevant alternatives are just those alternatives that
are true in worlds that are located at a certain position in this ordering. In order to assess this
suggestion, we need an account of this ordering. So: what does it take for a world to be
epistemically nearby? If all and only the nearby possible worlds are worlds in which the relevant
alternatives are true, then the question of what constitutes the nearby possible worlds is no
different than the question of what makes an alternative relevant. The only difference is that
adopting a nearby possible worlds framework suggests an answer in terms of an ordering and a
threshold, where an answer in those terms is not implicit in the relevant alternatives framework
more generally. So we might adopt a “nearby possible worlds” framework for thinking about
relevant alternatives, but this will not take us any closer to a substantive account of the relevant
alternatives.
As it turns out, several of the accounts to be offered from here on out do fit the model of
an ordering and a threshold. So if you are so inclined, you can read these accounts offered below
of the form “An alternative is relevant if and only if X” as “An alternative is relevant if and only
if it is true in an epistemically nearby possible world, where a proposition is true in an
epistemically nearby possible world if and only if X.” It makes no difference; the modal
framework is window dressing.
In sum, if we are taking relevant alternatives to be those alternatives that are true in
metaphysically or nomologically nearby possible worlds as the traditional account suggests, our
account will get the wrong results. Our account of what it takes to be a relevant alternative must
be able to deal with necessary truths and falsehoods in a substantive way. If, as is more plausible,
we mean that relevant alternatives just are those alternatives that are true in epistemically nearby
156

possible worlds, we've really only settled on a formal framework for thinking about relevance.
But a formal framework for thinking about relevance isn't an account of relevance. So the
traditional account is no help in finding what makes an alternative relevant. We need something
better.
2.3 Skeptical Hypotheses
The nearby possible worlds conception turned out to be either false or lacking in
substance. What this suggests is that we look for a substantive view that rules out all of the
alternatives that we need to rule out. And maybe we've been over-thinking things when it comes
to finding a substantive view that rules out the skeptical hypotheses that we need to rule out in
order to make this anti-skeptical strategy work. Maybe the right thing to say is that an alternative
is irrelevant just in case it's a skeptical hypothesis! This is surely substantive, and it's guaranteed
to rule out any skeptical hypothesis that we might have to deal with, including both moral
skeptical hypotheses like the Natural World hypothesis and external world skeptical hypotheses.
So the view that all skeptical hypotheses are irrelevant may seem perfectly formulated to get the
results we want in every case.
Unfortunately, the view that all skeptical hypotheses are irrelevant radically
overgenerates irrelevance. To see why this is so, remember that a hypothesis is “skeptical” only
relative to some kind of discourse. Talk about brains in vats or evil demons are hypotheses that
are skeptical relative to our discourse about the external world, while talk about evolutionary
debunking explanations is a skeptical hypothesis relative to our moral discourse.
9
Yet there are
some discourses that are fundamentally flawed. Phlogiston discourse, witch discourse, astrology
discourse, and many other discourses that are (or were) taken seriously are just bad discourses.

9
Recall: a skeptical hypothesis consists of a Falsity Condition and an Appearance Condition. The Falsity
Condition states that some target proposition is false, and the Appearance Condition gives an explanation of why
it will seem to the subject that the proposition is true, in a way that is consistent with the Falsity Condition.
157

They are things that we should be skeptical about – the only thing we should be saying about
these subjects is that they don’t exist. Yet any alternative to these discourses would count as a
skeptical alternative relative to these discourses – and thus, by the standard under consideration,
would be irrelevant.
To illustrate, recall the example of the Copernican Skeptical Hypothesis from Chapter 1.
According to the Copernican Skeptical Hypothesis, the common-sense belief that the sun
revolves around the Earth is false, but the rotation of the Earth causes it to look to people on
Earth as though the sun revolves around the Earth. The Copernican Skeptical Hypothesis has
what it takes to be a skeptical hypothesis. But it's not an irrelevant alternative. It's precisely the
kind of alternative that one must consider when attempting to decide whether it's true that the sun
revolves around the Earth. Similar considerations apply to any of the flawed discourses of the
previous paragraph. There is no such thing as witches or phlogiston, and so we can formulate
alternatives to these hypotheses that are skeptical relative to those discourses, but which don't
seem crazy or outlandish. In fact, at least one of the alternatives that we can devise to these
defective discourses will be a statement of how things really are.
The view that all skeptical hypotheses are irrelevant means that any scenario that meets
the criteria for being a skeptical alternative need not be considered. But many scenarios can meet
these criteria, since skeptical hypotheses are defined in terms of how they relate to some other
hypothesis. So some very sensible hypotheses will count as skeptical hypotheses, relative to
some very strange beliefs. We should not say these hypotheses are irrelevant just because they
are skeptical relative to strange beliefs. In fact, these hypotheses seem to be particularly relevant,
insofar as they are sensible corrections to strange beliefs. These are exactly the kinds of
hypotheses we should be considering when trying to figure out whether or not our bizarre beliefs
158

are really likely to be true.
The point here is an extension of the main point of Chapter 1. While we often think of a
skeptical hypothesis as being a remote, strange scenario like the brain-in-a-vat scenario, the
strangeness of this scenario is not a function of its being a skeptical hypothesis. Rather, it is a
function of its being a skeptical hypothesis relative to our beliefs about the external world.
Making a skeptical scenario relative to our external world beliefs requires us to get pretty
creative. No such creativity is required to devise a skeptical hypothesis relative to beliefs about
witches. So trying to identify the irrelevant alternatives by thinking about things in terms of
skeptical hypotheses is a step in the wrong direction. Skeptical hypotheses come in all shapes
and sizes. We should be trying to figure out what features of the brain-in-a-vat scenario make it
not worth considering.
2.4 What normal people think
What separates the irrelevant skeptical hypotheses (like the brain in a vat hypothesis)
from the relevant skeptical hypotheses (like the Copernican Skeptical Hypothesis)? One intuitive
thought is that the view that the Earth revolves around the sun is just more likely than the view
that I am a brain in a vat. This, as we will see, is a promising thought. But we need to make it
more precise.
One attempt to do this has been suggested by Patrick Rysiew (2001). According to
Rysiew, “the relevant alternatives are fixed by what we (normal) humans take to be the likely
counter-possibilities to what the subject is said to know” (Rysiew 2001, p. 488, italics in
original). This account of relevant alternatives seems to get us what we want. Normal humans do
not normally take radical skeptical hypotheses to be likely counter-possibilities. Hence, if the
relevant alternatives are determined by the standards of typical agents in typical circumstances,
159

skeptical alternatives will not be relevant alternatives.
Any account of relevant alternatives will state the alternatives that one must rule out in
order to have knowledge. So it would be extraordinarily convenient if the alternatives that we
need to rule out in order to have knowledge of some proposition just are the alternatives that we
do consider when we are trying to determine the truth of that proposition. But we're probably not
that lucky. It is rather likely that normal people will frequently consider alternatives that are
irrelevant, and almost certain that normal people fail to consider alternatives that are relevant.
10

We consider alternatives that are irrelevant whenever we want to hold onto hope that some
proposition for which we have overwhelming evidence is false, or when superstitions or various
other cognitive defects lead us to think through dead-end possibilities. For instance, there have
been societies in which the kinds of discourses we looked at in the previous section – witch
discourse, phlogiston discourse, etc. – are taken seriously. But an inclination towards superstition
should not make skeptical alternatives relevant. More general, we will often consider alternatives
that are, as Walter Sinnott-Armstrong puts it, too outlandish to be worth considering. But a
fascination with outlandish alternatives ought not raise our epistemic burdens.
We also frequently fail to take certain relevant alternatives into account in our reasoning.
Nearly every time we can rightly accuse someone of having made a hasty decision or “jumped to
a conclusion,” what we are accusing this person of is not considering a possible alternative that
really was relevant. Thus, we have good reason to believe that typical people ignore relevant
alternatives and consider irrelevant alternatives.
But even if typical human beings are, in fact, good judges of relevance, this can at most
serve to provide us with a heuristic as to what alternatives are relevant; it cannot be our account

10
Walter Sinnott-Armstrong rejects the view that the relevant alternatives are determined by “the usual epistemic
standards” - a view that is very similar (if not identical) to Rysiew's view – for the same reasons. He attributes
this point originally to Fogelin (1994).
160

of what alternatives are relevant. This is because what normal people consider and what counts
as relevant have different modal profiles. Consider a possible world in which normal people have
a brain defect that makes them fail to consider propositions about people whose names begin
with “K.” We're throwing a party, and Karen is coming
11
, although I'm not considering her at the
moment because of her first name. If I claim to know that we can serve pigs in a blanket at the
party since none of the guests are vegetarians (having arrived at this conclusion after thinking of
all the guests except for Karen, who is luckily not a vegetarian), the fact that I have made the in-
this-world-all-too-common mistake of forgetting about Karen doesn't mean that I do know that
none of the guests are vegetarians. I should have thought of Karen – I would need to in order to
know that none of the guests are vegetarians – but I didn't. Consideration of worlds like this
shows us that being considered by normal people can't be what makes an alternative relevant.
2.5 Bayesianism
Asking what typical people think doesn't give us a good account of what it takes for a
proposition to be likely, or worth considering. Perhaps we should understand this notion of
likelihood probabilistically. This suggests a Bayesian approach. On this approach, what
determines whether an alternative is relevant or not is just the subject's prior credence in that
alternative. More formally, there exists some threshold, t, such that for any alternative A, A is
relevant if and only if Pr(A) > t. I think there is something right about this. The alternatives that
we should be ruling out as irrelevant are those alternatives that are so crazy that it's not worth
considering them at all. And a low prior probability might look like the right way to capture this
thought. But sadly, this will not do. People can maintain an irrationally low credence in any
proposition, regardless of what evidence they have at hand.
Perhaps this modification will help: there exists some threshold, t, such that for any

11
Karen's parents must have been abnormal by the standards of this world to even think of naming her “Karen.”
161

subject S and alternative A, A is relevant if and only if Pr(A) > t, and S is rational (given his
evidence) to have a credence of Pr(A) in A. This is better, but it still will not do. Jonathan V ogel
(1999) has given a critique of any kind of probability-over-a-threshold view, including both
objective and subjective Bayesianism.
12
If a probability-over-a-threshold view is correct, then
we can easily construct a procedure for rendering any alternative whatsoever relevant. Take any
proposition that has a probability over the threshold – call this alternative R. Then take the
disjunction of R and any proposition whatsoever, I. The resulting disjunction, R v I, will have a
probability above the threshold because R does. But then, in order to have knowledge, one must
be in a position to rule out R v I, which involves being able to rule out I as well as R. So in order
to have knowledge, one must be able to rule out any proposition I – even if I is a skeptical
hypothesis. So, on this Bayesian account, the relevant alternatives view will collapse into the “all
alternatives” view. That's a devastating problem.
2.6 Taking Stock
Thus far, we have looked at and rejected a number of candidates for what makes an
alternative relevant. While we do not yet have an account of relevant alternatives in hand, we
have identified a number of criteria that any adequate account of relevant alternatives should
meet.
First, our account must be Anti-Skeptical: it must be able to rule as irrelevant the
possibility that we are being deceived by evil demons or that we are brains in vats. This, after all,
is the whole point of putting forward a relevant alternatives account of knowledge – to be able to
ignore those skeptical alternatives that we can properly ignore, yet are unable to rule out. Second,
it must be Limited – it cannot be so strong that it rules as irrelevant skeptical hypotheses to
deficient discourses, like witch discourse. The view that all skeptical hypotheses are irrelevant

12
His target is, specifically, Cohen (1988), who puts forward a probabilistic view.
162

fails because it does not meet this criterion. Third, it must be Broadly Applicable, by which I
mean that it must be able to yield a set of relevant alternatives for necessary truths. Since many
of our moral beliefs are beliefs in (purportedly) necessary truths, if our account is to be helpful in
adjudicating which alternatives to moral propositions are relevant, our account cannot be
restricted to the realm of contingent truths. Fourth, it must be possible for disjunction
introduction to turn a relevant alternative into an irrelevant alternative. Call accounts that meet
this fourth criterion Non-Extendable.

3. Relevant Alternatives and Empirical Virtues
In this section, I will present an account of what makes an alternative relevant that meets
our criteria and so represents, I believe, our best chance of coming up with an adequate account
of relevant alternatives.
The basic idea behind the account I'll offer up here is one that we've encountered already,
but is worth bringing into more narrow focus. Rysiew claimed that the relevant alternatives are
those that are considered by us to be likely to be true. The interpretation of the “nearby possible
worlds” account on which we take the relevant “nearness” metric to be worlds that are
epistemically nearby suggested that the alternatives that we should count as relevant must
actually be likely, given our evidence. And Walter Sinnott-Armstrong has characterized the
irrelevant alternatives as outlandish. They are just too crazy to be considered, and it's precisely in
virtue of this fact that these are alternatives that need not be considered. I think this is a
promising thought: an alternative is irrelevant if and only if it is outlandish. But this thought
needs more refinement. What does it take for a hypothesis to be outlandish?
We can get a grip on the concept of “outlandishness” by focusing on the Anti-Skeptical
163

criterion for relevant alternatives. According to this criterion, our account of the relevant
alternatives must rule as irrelevant evil demons and brains in vats. Why call these theories
outlandish? The proposal that we are being deceived by an evil demon might be an important
possibility in virtue of its ramifications for the possibility of knowledge. But it’s a terrible theory
of why we actually have the experiences that we have. While there are a significant number of
skeptics who deny the possibility of knowledge in light of the possibility of evil demons, no
serious philosopher holds that evil demons are what actually explain our visual experiences. This
seems rational. As we saw in Chapter 5, there is a strong case to be made that explanations of our
visual experiences in terms of brains in vats and evil demons just are not particularly good
explanations of those visual experiences, at least from the standpoint of the empirical virtues of
simplicity, and unified explanatory power. The hypothesis that there is an external world explains
our visual experiences in a much more elegant way than an ad hoc story about the whims of evil
demons.
Thus, a very natural way to think about outlandish skeptical hypotheses is as skeptical
hypotheses where the Appearance Condition explains our internally-accessible evidence in a way
that does not manifest the empirical virtues. And that, I believe, is what makes an alternative
relevant: An alternative is relevant if and only if it manifests the empirical virtues to some
minimal extent, where the empirical virtues are just the factors in virtue of which an explanation
counts as a good explanation (simplicity, fit to data, no ad hoc explanations, etc.). Notice here
that an alternative is relevant just in case it is a good explanation. It need not be the best
explanation of anything. What marks the difference? All of the factors that make an explanation
“the best explanation” are scalar to a certain degree. A hypothesis can be more or less simple, it
can fit our data to a greater or lesser extent, it can provide better or worse higher-level
164

explanations for lower-level propositions (this is what it takes to be more or less ad hoc). We
weight and aggregate these factors to come up with a scale on which some explanations are
better than others. When we talk about the best explanation, we are talking about the hypothesis
that scores highest on this scale out of all the possible hypotheses available to us. But when I
speak here of a hypothesis that manifests the empirical virtues to some minimal extent, what I
mean is not the hypothesis that scores the highest, but rather any hypothesis that ranks above a
certain threshold. In other words,
Relevance as Empirical Virtue (REV): If the Appearance Condition of a
skeptical hypothesis explanation manifests the empirical virtues to a great
enough extent to be worth considering, then that skeptical hypothesis is a
relevant alternative.
REV is a natural thought about what could make a hypothesis relevant. It also meets our
criteria from Section 2.6 for what we want out of an account of relevant alternatives. First, it is
Anti-Skeptical. As we saw in the previous chapter, hypotheses about evil demons or brains in
vats do fail along several of the dimensions that make a given hypothesis the best explanation.
Second, REV is Limited, in that it does not rule every skeptical hypothesis to be irrelevant. The
Copernican Skeptical Hypothesis is a model of empirical virtue. This is why it is a relevant
alternative to views according to which the Earth is located at the center of the cosmos. Third,
REV is Broadly Applicable, since it can offer up substantive contrast classes for necessary truths.
Consider the identity of water and H2O. At a certain point in history, before Lavoisier had
performed his experiments, the hypothesis that water was an element was just as good an
explanation of the observed behavior of water as the hypothesis that water is H2O – this was just
because we didn't have the technology or wherewithal to construct experiments which would
165

give us data that needed to be explained. At this time – or, more relevantly, relative to this body
of evidence – there were a number of relevant alternatives to the hypothesis that water is H2O.
And fourth, this account is Non-Extendable. If an explanation E manifests the empirical virtues
to some minimal extent, it does not follow that, for all propositions I, E˅I will be a good
explanation of anything. Arbitrary disjunction introduction makes explanations less simple and
unified.
One potential objection says that what I've provided isn't really a “relevant alternatives”
account; I've just reintroduced IBE. This objection is misguided. I haven't just reintroduced IBE,
since my account makes no mention of the best explanation, but only talks about explanations
that are higher than some threshold. If there are multiple explanations above that threshold (as is
likely to be the case), some other method must be used to “rule out” competing relevant
alternatives. The method we use here might actually be IBE itself, but it might be something else
– coherence, support from foundational beliefs, reliability, or what have you.
The objector might press here by saying that while this method differs from IBE in its
general approach to knowledge, it is fundamentally no different than IBE when it comes to its
solution to the skeptical problem. I have two responses to this. First, I'm happy to bite the bullet
on this point, and see no reason why it would be troubling to do so. My goal here was to provide
an adequate account of the relevant alternatives anti-skeptical strategy, not to provide such an
account that is fundamentally different from the IBE strategy. That IBE and relevant alternatives
amount to the same thing as far as skepticism is concerned is an interesting result, but I see no
reason to think that it is problematic. Second, although this account of relevant alternatives has
followed the IBE anti-skeptical strategy very closely, they are not identical. The IBE strategy
says that we can rule out certain skeptical hypotheses in virtue of their not being the best
166

explanation of our experiences. The relevant alternatives account offered here, on the other hand,
is not committed to the possibility that IBE can rule out skeptical hypotheses. Rather, the
relevant alternatives account offered here says that, in virtue of failing along several dimensions
that happen to have explanatory relevance, skeptical hypotheses need not be ruled out. This kind
of view might appeal to someone who has doubts about whether or not IBE can work to “rule
out” skeptical alternatives,
13
but who still thinks that the factors relevant to explanation will
prove helpful in solving the skeptical challenge.

4. The Relevance of Moral Skepticism
If everything I've said to this point has been persuasive, I have little work left to do in
order to show that the Natural World hypothesis are relevant. As I argued in the last chapter,
Equal Moral Explanations is true – the Natural World hypothesis provides just as good (if not
better) explanations as moral hypotheses. Thus, when it comes to the matter of whether or not the
Natural World hypothesis is above the “minimal explanatory virtues” threshold articulated in the
last section, the answer must be: the Natural World hypothesis is above the threshold if moral
hypotheses are. And since moral hypotheses are above the threshold, the Natural World
hypothesis will be as well.
Therefore, the Natural World hypothesis
is relevant, and so the Relevant
Alternatives anti-skeptical strategy won't constitute an adequate answer to the Moral Closure
Argument.
Given that Equal Moral Explanations is true, the only way to argue that the Natural World
hypothesis is below the threshold of relevance would be to hold that moral beliefs are themselves
below the threshold of relevance. But this is not a move that an anti-skeptic would want to make;

13
For instance, if “ruling out” requires reducing the probability of the alternative to zero, IBE will be unable to
“rule out” any alternatives; IBE shows that certain alternatives are very likely, yet it is always possible that a less
simple alternative (e.g.) will be the correct one.
167

if moral propositions are themselves below the threshold, then one can be quite easily justified in
believing in moral nihilism, since any alternatives to nihilism will be irrelevant. That's not the
kind of concession a moral realist should want to make.
At this point, I conclude that the most reasonable account of what makes an alternative
relevant is the view that all alternatives that manifest the empirical virtues to some minimal
extent are relevant, and because of this, the Natural World hypothesis is relevant. And it should
not be surprising that the Natural World hypothesis should count as relevant where evil demons
and envatted brains do not. One of the main contentions of this dissertation is that moral
skepticism is an entirely reasonable position, because the skeptical alternatives that the moral
skeptic presents are not outlandish theories. That our moral beliefs are simply the product of
evolutionary and social pressures and not responses to an independent moral truth makes sense.
It's not a theory for cranks and madmen – it's a theory that a realist needs to engage with. In
short, it's a theory that's relevant.

5. Conclusion
I will close with a few comments on what I take the main lessons of the dissertation to be.
First, not all skeptical arguments are created equal. Some skeptical arguments are bound to be
both sound and compelling. Some skeptical arguments are not. Whether a particular skeptical
argument is compelling will depend essentially on the subject matter of the argument. Thus, there
is no reason to think that a Closure Argument for Moral Skepticism will be a bad argument
before assessing this argument on its merits. And, on inspection, we can see that a Moral Closure
Argument has a lot going in its favor.
Second, the moral non-naturalist faces some distinct and difficult epistemic challenges
168

that all stem from the fact that, for the non-naturalist, moral facts are not part of the causal order.
We’ve seen that a number of skeptical challenges can be answered by giving reasons to think that
some fact is located at a proper place in a complete explanation of why an agent has the belief in
question. Moral naturalists can, in principle, say that moral facts can play this explanatory role.
Moral non-naturalists cannot. And most of the epistemic problems for non-naturalism spring
from this fact.
Third, explanation plays a much larger role in giving an account of knowledge than many
have been willing to recognize. In contemporary epistemology, explanatory connections are
often set to the side in favor of modal connections – a contemporary epistemologist is much
more likely to give an account of knowledge in terms of what would have been true in nearby
possible worlds in which S believes that P than an account in terms of what explains S’s belief
that P. But this modal move is highly problematic and invariably generates incorrect results when
confronted with cases involving necessary truths. We’ve seen that modal accounts of knowledge
always end up sputtering when confronted with the question of how it could be both possible to
know that water is H2O, and difficult to know this fact. A shift to considering explanatory
connections helps to solve this problem by retaining what is intuitive about modal accounts of
knowledge. Formulating our account in categorical rather than counterfactual terms avoids a
number of the formal problems that face counterfactual accounts.
And finally, we’ve seen the importance of various manifestations of the Gettier problem
to the question of whether or not skeptical challenges can be answered. It is fairly easy to give an
account of a justified, true, belief in some proposition. One need only hold that the proposition is
true, show that this is the kind of thing one can believe, and then situate an agent in a
hypothetical situation where all of her evidence points overwhelmingly in favor of that
169

proposition’s being true. Accordingly, it’s possible to construct a case where one has a justified,
true belief in any proposition (for which a case can be constructed in which that proposition is
true). Knowledge, however, is substantially more demanding than mere justified, true belief. To
give an account of moral knowledge, an anti-skeptic must begin to tackle the Gettier problem to
offer an account of how our moral beliefs can be more than merely subjectively rational. But the
currents in the contemporary literature on the Gettier problem drag us away from the conclusion
that knowledge of non-natural moral truths is possible. Moral epistemology can benefit a great
deal by looking closer at the contemporary literature on the Gettier problem. I hope I have
advanced that conversation here.
170

Bibliography
Audi, R. (2004). The good in the right: A theory of intuition and intrinsic value. Princeton
University Press.
----- (2008). Intuition, inference, and rational disagreement in ethics. Ethical Theory and Moral
Practice 11 (5):475 - 492.
----- (2013). Moral perception. Princeton University Press.
Armstrong, D. M. (1973). Belief, Truth and Knowledge. London,Cambridge University Press.
Bedke, Matthew S. (2009). Intuitive non-naturalism meets cosmic coincidence. Pacific
Philosophical Quarterly 90 (2):188-209.
Benacerraf, Paul (1973). Mathematical truth. Journal of Philosophy 70 (19):661-679.
Berker, Selim (2014). Does Evolutionary Psychology Show That Normativity Is Mind-
Dependent? In Justin D'Arms & Daniel Jacobson (eds.), Moral Psychology and Human
Agency: Philosophical Essays on the Science of Ethics. Oxford University Press. 215-252.
BonJour, Laurence & Sosa, Ernest (2003). Epistemic Justification: Internalism Vs. Externalism,
Foundations Vs. Virtues. Blackwell Pub..
Boolos, George ; Burgess, John ; P., Richard & Jeffrey, C. (2007). Computability and Logic.
Cambridge University Press.
Brandt, R. B. (1954). Hopi Ethics: A Theoretical Analysis. University of Chicago Press.
Brink, David Owen (1989). Moral Realism and the Foundations of Ethics. Cambridge University
Press.
Burge, Tyler (2003). Perceptual entitlement. Philosophy and Phenomenological Research 67
(3):503-548.
Clarke-Doane, Justin (forthcoming). Moral Epistemology: The Mathematics Analogy. Noûs 48
171

(2):238-255.
Chisholm, Roderick M. (1977). Theory of Knowledge (2
nd
ed). Prentice-Hall.
Cohen, Stewart (1988). How to be a fallibilist. Philosophical Perspectives 2:91-123.
Comesaña, Juan (2006). A well-founded solution to the generality problem. Philosophical
Studies 129 (1):27 - 47.
Copp, David (2008). Darwinian skepticism about moral realism. Philosophical Issues 18
(1):186-206.
Cuneo, Terence (2006). Moral facts as configuring causes. Pacific Philosophical Quarterly 87
(2):141–162.
Cuneo, Terence & Shafer-Landau, Russ (2014). The moral fixed points: new directions for moral
nonnaturalism. Philosophical Studies 171 (3):399-443.
Dancy, Jonathan (2006). Nonnaturalism. In Oxford Handbook of Ethical Theory, ed. David
Copp. Oxford University Press.
DePaul, Michael (2009). Pyrrhonian moral skepticism and the problem of the
criterion. Philosophical Issues 19 (1):38-56.
DeRose, Keith (1995). Solving the skeptical problem. Philosophical Review 104 (1):1-52.
Dretske, Fred I. (1970). Epistemic operators. Journal of Philosophy 67 (24):1007-1023.
Elga, Adam (2007). Reflection and disagreement. Noûs 41 (3):478–502.
Enoch, David (2011). Taking Morality Seriously: A Defense of Robust Realism. Oxford
University Press.
Feldman, Richard (1988). Having evidence. In D. F. Austin (ed.), Philosophical Analysis.
Kluwer Academic Publishers. 83--104.
----- (2003). Epistemology. Pearson College Division.
172

Field, Hartry (1989). Realism, Mathematics & Modality. Basil Blackwell.
Fitzpatrick, William (2008). Robust ethical realism, non-naturalism, and normativity. Oxford
Studies in Metaethics 3:159-205.
Rysiew, Patrick (2001). The context-sensitivity of knowledge attributions. Noûs 35 (4):477–514.
Gettier, Edmund (1963). Is Justified True Belief Knowledge? Analysis 23 (6):121-123.
Harman, Gilbert (1977). The Nature of Morality: An Introduction to Ethics. Oxford University
Press.
Harman, Gilbert, and Judith Jarvis Thomson. Moral relativism and moral objectivity. Oxford:
Blackwell, 1996.
Heller, Mark (1999). Relevant alternatives and closure. Australasian Journal of Philosophy 77
(2):196 – 208.
Huemer, M. (2001). Skepticism and the Veil of Perception. Rowman & Littlefield.
----- (2005). Ethical Intuitionism. Palgrave Macmillan.
Joyce, Richard (2006). The Evolution of Morality. MIT Press.
----- (forthcoming) Reply: Confessions of a Modest Debunker. In Explanation in Mathematics
and Ethics, ed. U Leibowitz and N Sinclair. Oxford University Press.
Kahneman, D. (2011). Thinking, fast and slow. Macmillan.
Kotzen, Matthew (manuscript). A Formal Account of Epistemic Defeat.
Kulp, Christopher B. (2011). Moral facts and the centrality of intuitions. In Jill Graper
Hernandez (ed.), The New Intuitionism. 48--66.
Lasonen-Aarnio, M. (2010). Unreasonable knowledge. Philosophical Perspectives,24(1), 1-21.
Lehrer, Keith (1974). Knowledge. Clarendon Press.
Lehrer, Keith & Paxson Jr, Thomas (1969). Knowledge: Undefeated justified true belief. Journal
173

of Philosophy 66 (8):225-237.
Lewis, David (1996). Elusive knowledge. Australasian Journal of Philosophy 74 (4):549 – 567.
Lutz, Matt (2014). The pragmatics of pragmatic encroachment. Synthese:1-24.
Maguire, Barry (forthcoming). Grounding the Autonomy of Ethics. Oxford Studies in Metaethics
(Vol. 10).
Majors, Brad (2007). Moral explanation. Philosophy Compass 2 (1):1–15.
McCann, Hugh J. (2011). Conative intuitionism. In Jill Graper Hernandez (ed.), The New
Intuitionism. 48--66.
McGrath, Sarah (2004). Moral knowledge by perception. Philosophical Perspectives 18 (1):209–
228.
----- (2014). Relax? Don’t do it! Why moral realism won’t come cheap. In Oxford Studies in
Metaethics 9 (2014): 186-214.
Moore, G. E. (1903). Principia Ethica. Dover Publications.
Nelson, Mark T. (1995). Is it Always Fallacious to Derive Values From Facts? Argumentation 9
(4):553-562.
----- (2003). Sinnott–Armstrong's moral scepticism. Ratio 16 (1):63–82.
----- (2006). The Possibility of Inductive Moral Arguments. Philosophical Papers 35 (2):231-
246.
Nozick, Robert (1981). Philosophical Explanations. Harvard University Press.
Oddie, Graham (2005). Value, Reality, and Desire. Clarendon Press.
Olson, Jonas (2014). Moral Error Theory: History, Critique, Defence. Oup Oxford.
Parfit, Derek (2011). On What Matters. Oxford University Press.
Pigden, Charles R. (1989). Logic and the autonomy of ethics. Australasian Journal of
174

Philosophy 67 (2):127 – 151.
Pollock, John (1987). Defeasible Reasoning. Cognitive Science 11 (4):481-518.
Pollock, John L. & Gillies, Anthony S. (2000). Belief revision and epistemology. Synthese 122
(1-2):69-92.
Prichard, H. A. (1912). Does moral philosophy rest on a mistake? Mind 21 (81):21 - 37.
Prior, A. N. (1960): 'The Autonomy of Ethics', Australiasian Journal of Philosophy 38.
Pritchard, Duncan (2010). Relevant alternatives, perceptual knowledge and
discrimination. Noûs 44 (2):245-268.
Pryor, James (2000). The skeptic and the dogmatist. Noûs 34 (4):517–549.
Putnam, Hilary (1981). Reason, Truth and History. Cambridge University Press.
Ross, Jacob & Schroeder, Mark (2014). Belief, Credence, and Pragmatic
Encroachment. Philosophy and Phenomenological Research 88 (2):259-288.
Ruse, Michael (1986). Taking Darwin Seriously: A Naturalistic Approach to Philosophy.
Prometheus Books.
Rysiew, Patrick (2001). The context-sensitivity of knowledge attributions. Noûs 35 (4):477–514.
Schafer, Karl (2014). Knowledge and Two Forms of Non-Accidental Truth. Philosophy and
Phenomenological Research 89 (2):373-393.
Shafer-Landau, Russ (2003). Moral Realism: A Defence. Oxford University Press.
Scanlon, T. M. (2014). Being Realistic About Reasons. Oxford University Press.
Schroeder, Mark (2011). Holism, Weight, and Undercutting. Noûs 45 (2):328 - 344.
----- (forthcoming). ‘Knowledge is Belief for Sufficient (Objective and Subjective) Reason.’
Oxford Studies in Epistemology, volume 5.
Searle, John R. (1964). How to derive "ought" from "is". Philosophical Review 73 (1):43-58.
175

Setiya, Kieran (2012). Knowing Right From Wrong. Oxford University Press.
Shoemaker, Sydney (1963). Self-Knowledge and Self-Identity. Cornell University Press.
Sinnott-Armstrong, Walter (1996). Moral Skepticism and Justification. In Walter Sinnott-
Armstrong & Mark Timmons (eds.), Moral Knowledge? New Readings in Moral
Epistemology. Oxford University Press.
----- (2006). Moral Skepticisms. Oxford University Press.
Sosa, Ernest (1999). How to defeat opposition to Moore. Philosophical Perspectives 13
(s13):137-49.
Sripada, Chandra (2008). Nativism and Moral Psychology: Three Models of the Innate Structure
that Shapes the Contents of Moral Norms. In Moral Psychology, ed. Walter Sinnott-
Armstrong. MIT Press.
Stampe, Dennis W. (1987). The authority of desire. Philosophical Review 96 (July):335-81.
Star, Daniel (2008). Moral knowledge, epistemic externalism, and intuitionism. Ratio 21 (3):329-
343.
Stevenson, Charles (1937). The emotive meaning of ethical terms. Mind 46 (181):14-31.
Stine, G. C. (1976). Skepticism, relevant alternatives, and deductive closure.Philosophical
Studies, 29(4), 249-261.
Street, Sharon (2006). A Darwinian dilemma for realist theories of value. Philosophical Studies
127 (1):109-166.
----- (ms). Objectivity and Truth: You'd Better Rethink It.
Sturgeon, Scott (2008). Reason and the grain of belief. Noûs 42 (1):139–165.
Sturgeon, Nicholas (1988). Moral Explanations. In Essays on Moral Realism, ed. Geoffrey
Sayre-McCord. Cornell University Press.
176

Thomson, Judith Jarvis (1990). The Realm of Rights. Harvard University Press.
Thurow, Joshua C. (2013). The defeater version of Benacerraf's problem for a priori
knowledge. Synthese 190 (9):1587-1603.
Tropman, Elizabeth (2011). Non-Inferential Moral Knowledge. Acta Analytica 26 (4):355-366.
----- (2014). Evolutionary debunking arguments: moral realism, constructivism, and explaining
moral knowledge. 17 (2):126-140.
Unger, Peter (1968). An analysis of factual knowledge. Journal of Philosophy 65 (6):157-170.
Vavova, Katia (2014a). Debunking Evolutionary Debunking. Oxford Studies in Metaethics 9:76-
101.
----- (2014b). Moral disagreement and moral skepticism. Philosophical Perspectives, 28(1), 302-
333.
V ogel, Jonathan (1990). Cartesian skepticism and inference to the best explanation. Journal of
Philosophy 87 (11):658-666.
----- (1999). The new relevant alternatives theory. Philosophical Perspectives 13 (s13):155-180.
Wedgwood, Ralph (2007). The Nature of Normativity. Oxford University Press.
White, Roger (2010). You just believe that because…. Philosophical Perspectives 24 (1):573-
615.
Wielenberg, Erik J. (2010). On the evolutionary debunking of morality. Ethics 120 (3):441-464.
Williamson, Timothy (2000). Knowledge and its Limits. Oxford University Press.
Wright, Crispin (1983). Frege's Conception of Numbers as Objects (Vol. 2). Aberdeen University
Press.
Yamada, Masahiro (2011). Getting It Right By Accident. Philosophy and Phenomenological
Research 83 (1):72-105.
177

Zimmerman, Aaron (2010). Moral Epistemology. Routledge.

Abstract (if available)

Abstract In this dissertation, I examine the question of whether we have particular reason to be skeptical about moral facts, or whether the best arguments for moral skepticism are just special cases of more general skeptical concerns. I argue that, although moral skeptical arguments can be given that share structural similarities with other, more general skeptical arguments, anti-skeptics about morality lack resources to respond to these challenges that anti-sketpics in other areas (for instance, about the external world) lack. When we closely scrutinize the ways in which skeptical arguments operate, and the ways in which those skeptical arguments can be countered, the case for moral skepticism emerges.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Positivist realism

PDF

A deontological explanation of accessibilism

PDF

Contrastive reasons

PDF

Reasons, obligations, and the structure of good reasoning

PDF

Extending credit

PDF

Beliefs that wrong

PDF

A perceptual model of evaluative knowledge

PDF

Rationality and the primacy of the occurrent

PDF

Merely verbal disputes in philosophy

PDF

Process-oriented rationality

PDF

The mental states first theory of promising

PDF

Rethinking reductive realism in ethics

PDF

Reasoning with degrees of belief

PDF

The virtue of reasonableness: on the normative epistemic implications of public reason liberalism

PDF

Thomas Reid on singular thought

PDF

Language and the structure of Berkeley's world

PDF

Morality with Humean foundations: a proposal for a solution to the accommodation problem

PDF

Describing authority

PDF

Toward a more perfect liberalism: perfectionism in Kantian political philosophy

PDF

Units of agency in ethics

Asset Metadata

Creator Lutz, Matt (author)

Core Title The case for moral skepticism

School College of Letters, Arts and Sciences

Degree Doctor of Philosophy

Degree Program Philosophy

Publication Date 05/13/2015

Defense Date 03/10/2015

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag closure Arguments,evolutionary debunking,moral epistemology,OAI-PMH Harvest,skepticism

Format application/pdf (imt)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Finlay, Stephen (committee chair), Schroeder, Mark (committee member), Van Cleve, James (committee member), Walker, Rachel (committee member), Wedgwood, Ralph (committee member)

Creator Email lutzm@usc.edu,mattlutz326@gmail.com

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c3-567972

Unique identifier UC11300525

Identifier etd-LutzMatt-3447.pdf (filename),usctheses-c3-567972 (legacy record id)

Legacy Identifier etd-LutzMatt-3447.pdf

Dmrecord 567972

Document Type Dissertation

Format application/pdf (imt)

Rights Lutz, Matt

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA